# Real Alpha

> The best minutes from long interviews with the people building AI. Each moment is a few minutes of a podcast or YouTube interview with a founder, executive or researcher, with who said it, the claim, its takeaways and the transcript, playing on its own page.

Every page has a Markdown copy at its address plus index.md, like /s/<slug>/index.md, and answers Accept: text/markdown with it. The links below go to the copies; cite the page each copy names.

Cite a moment by its page. Each page names the speaker, the episode and the second the clip starts; episodes are at /v/<YouTube id>/, shows at /shows/<id>/, builders at /builders/<id>/ and everyone else quoted at /people/<id>/, each with what they said.

## Pages

- [Today's moments](https://realalpha.app/index.md): the newest moments, best first
- [Best AI podcasts](https://realalpha.app/best/ai-podcasts/index.md): the AI podcasts we follow, ranked from what we've cut from each
- [Best tech podcasts](https://realalpha.app/best/tech-podcasts/index.md): the tech podcasts we follow, ranked from what we've cut from each
- [Best business podcasts](https://realalpha.app/best/business-podcasts/index.md): the business podcasts we follow, ranked from what we've cut from each
- [Builders](https://realalpha.app/builders.md): the founders, executives and researchers whose interviews are cut, each with their moments
- [Companies](https://realalpha.app/companies/index.md): the AI companies they lead
- [Hosts](https://realalpha.app/hosts/index.md): the hosts of the shows we cut
- [Big questions](https://realalpha.app/q.md): the questions people ask about AI, each answered by the moments where builders answer it
- [MCP server](https://realalpha.app/mcp/index.md): connect an AI assistant to https://realalpha.app/mcp (Streamable HTTP, no sign-in) to search these moments by topic, person, company or show, and to read how builders answer the big questions
- [Predictions ledger](https://realalpha.app/predictions/index.md): every dated prediction AI leaders made in the moments we cut, with the quote, the deadline and what came of it. Of 23 predictions with deadlines, none has come due yet. 6 more have no deadline. As CSV, free to reuse with credit: https://realalpha.app/predictions.csv
- [Sitemap](https://realalpha.app/sitemap.xml): every page

## Shows

Each show has a page at /shows/<id>/, ranked as /best/ai-podcasts/ ranks them. Each moment we’ve cut from a show earns a point, plus one when an AI builder is the one talking and one when its episode cleared our views bar, and a moment’s points halve every 90 days after its episode came out.

- [MTS](https://realalpha.app/shows/mts/index.md): No. 1. MTS is an interview show with the people building AI, hosted by Theo Jaffee and Sophia Dew, with a new episode every day. 33 moments cut from 12 episodes, 9 with an AI builder talking, mostly on AI agents and safety & alignment. Top guests: Nenad Tomašev, Tiago Sada, Bernardo Aceituno, Aidan Smith.
- [Latent Space](https://realalpha.app/shows/latent-space/index.md): No. 2. Latent Space is an interview show with the people building AI, hosted by Shawn Wang (swyx) and Alessio Fanelli, with a new episode every day. 22 moments cut from 8 episodes, 13 with an AI builder talking, mostly on enterprise AI and robotics. Top guests: Erik Bernhardsson, Ian Quigley, Jason Ma, Liam Fedus.
- [All-In](https://realalpha.app/shows/all-in/index.md): No. 3. All-In is a tech news show, hosted by Chamath Palihapitiya, Jason Calacanis, David Sacks and David Friedberg, with a new episode a few times a week. 20 moments cut from 7 episodes, 5 with an AI builder talking, mostly on enterprise AI and AI business models. Top guests: Alex Karp, Arvind Krishna, Ben Horowitz, Billy Evans.
- [a16z Podcast](https://realalpha.app/shows/a16z/index.md): No. 4. a16z Podcast is an investing and business show, hosted by Erik Torenberg, with a new episode every day. 22 moments cut from 9 episodes, 11 with an AI builder talking, mostly on AI agents and chips & GPUs. Top guests: Alex Karp, Jad Naous, Nathan Lambert, Matt Garman.
- [Sourcery](https://realalpha.app/shows/sourcery/index.md): No. 5. Sourcery is an interview show with the people building AI, hosted by Molly O'Shea, with a new episode a few times a week. 14 moments cut from 5 episodes, 8 with an AI builder talking, mostly on AI agents and building AI products. Top guests: Guillermo Rauch, Alex Karp, Aravind Srinivas, Jim Lanzone.
- [Lex Fridman Podcast](https://realalpha.app/shows/lex-fridman/index.md): No. 6. Lex Fridman Podcast is a long-form conversation show on research and big ideas, hosted by Lex Fridman, with a new episode every two weeks. 29 moments cut from 12 episodes, 5 with an AI builder talking, mostly on AI agents and coding agents. Top guests: Jensen Huang, Peter Steinberger, Gary W. Gallagher, Don Lincoln.
- [Sources](https://realalpha.app/shows/sources/index.md): No. 7. Sources is a tech news show, hosted by Alex Heath, with a new episode about twice a week. 9 moments cut from 3 episodes, 9 with an AI builder talking, mostly on open-weight models and inference & cost. Top guests: Sundar Pichai, Misha Laskin, Cristiano Amon, Ioannis Antonoglou.
- [Lenny's Podcast](https://realalpha.app/shows/lennys-podcast/index.md): No. 8. Lenny's Podcast is an interview show with the people building AI, hosted by Lenny Rachitsky, with a new episode every week. 9 moments cut from 3 episodes, 9 with an AI builder talking, mostly on building AI products and AI startups. Top guests: Emmanuel Ameisen, Josh Woodward, Tibo Sottiaux.
- [Training Data](https://realalpha.app/shows/training-data/index.md): No. 9. Training Data is an interview show with the people building AI, hosted by Sonya Huang and Pat Grady, with a new episode every week. 15 moments cut from 5 episodes, 6 with an AI builder talking, mostly on training data and talent & hiring. Top guests: David Silver, Ali Ghodsi, Josh Gao, Luca Ferrari.
- [Moonshots](https://realalpha.app/shows/moonshots/index.md): No. 10. Moonshots is a tech news show, hosted by Peter Diamandis, Salim Ismail, Dave Blundin and Alexander Wissner-Gross, with a new episode a few times a week. 9 moments cut from 3 episodes, 1 with an AI builder talking, mostly on data centers & energy and chips & GPUs. Top guests: Emad Mostaque, Chamath Palihapitiya, Anousheh Ansari.
- [The Knowledge Project](https://realalpha.app/shows/knowledge-project/index.md): No. 11. The Knowledge Project is a long-form conversation show on research and big ideas, hosted by Shane Parrish, with a new episode every two weeks. 13 moments cut from 5 episodes, mostly on investing in AI and AI startups. Top guests: Ryan Serhant, Tobi Lütke, Brad Jacobs, Gio Valiante.
- [Founders](https://realalpha.app/shows/founders/index.md): No. 12. Founders is an investing and business show, hosted by David Senra, with a new episode a few times a week. 12 moments cut from 4 episodes, 3 with an AI builder talking, mostly on big Tech's AI bets and consumer AI. Top guests: Alexandr Wang, Kurt MacAlpine, Ben Wilson, Ari Emanuel.
- [The Cognitive Revolution](https://realalpha.app/shows/cognitive-revolution/index.md): No. 13. The Cognitive Revolution is a long-form conversation show on research and big ideas, hosted by Nathan Labenz, with a new episode a few times a week. 11 moments cut from 4 episodes, 3 with an AI builder talking, mostly on biology & drugs and safety & alignment. Top guests: Woodson Martin, Alex Reeves, Alan Chan, Eric Bigelow.
- [The MAD Podcast](https://realalpha.app/shows/mad-podcast/index.md): No. 14. The MAD Podcast is an interview show with the people building AI, hosted by Matt Turck, with a new episode every week. 9 moments cut from 3 episodes, 6 with an AI builder talking, mostly on data & databases and reasoning & RL. Top guests: Pierre Stock, Diogo Almeida, Andy Pavlo.
- [Cheeky Pint](https://realalpha.app/shows/cheeky-pint/index.md): No. 15. Cheeky Pint is an interview show with the people building AI, hosted by John Collison, with a new episode every week. 8 moments cut from 3 episodes, 2 with an AI builder talking, mostly on data centers & energy and robotics. Top guests: Alexandr Wang, Mark Zuckerberg, Toto Wolff, Will O’Brien.
- [Big Technology Podcast](https://realalpha.app/shows/big-technology/index.md): No. 16. Big Technology Podcast is a tech news show, hosted by Alex Kantrowitz, with a new episode a few times a week. 6 moments cut from 2 episodes, 3 with an AI builder talking, mostly on chips & GPUs and data centers & energy. Top guests: Sundar Pichai, Dylan Patel, Jordan Nanos.
- [Unsupervised Learning](https://realalpha.app/shows/unsupervised-learning/index.md): No. 17. Unsupervised Learning is an interview show with the people building AI, hosted by Jacob Effron, with a new episode every week. 11 moments cut from 4 episodes, 11 with an AI builder talking, mostly on reasoning & RL and enterprise AI. Top guests: Yash Patil, Yann LeCun, Jakub Pachocki, Buck Shlegeris.
- [Behind the Craft](https://realalpha.app/shows/behind-the-craft/index.md): No. 18. Behind the Craft is an interview show with the people building AI, hosted by Peter Yang, with a new episode every week. 9 moments cut from 3 episodes, 3 with an AI builder talking, mostly on AI agents and coding agents. Top guests: Sam Stephenson, Ras Mic, Lauren Tan, Peng Zheng.
- [BG2](https://realalpha.app/shows/bg2/index.md): No. 19. BG2 is an investing and business show, hosted by Brad Gerstner and Bill Gurley, with a new episode every three weeks. 17 moments cut from 6 episodes, 8 with an AI builder talking, mostly on data centers & energy and investing in AI. Top guests: Jensen Huang, Satya Nadella, Andrew Fox, Gavin Baker.
- [Minus One](https://realalpha.app/shows/minus-one/index.md): No. 20. Minus One is an interview show with the people building AI, hosted by Aditya Agarwal, with a new episode every two weeks. 6 moments cut from 2 episodes, 6 with an AI builder talking, mostly on robotics and training data. Top guests: Sergey Levine, Dmitri Dolgov.
- [The Pragmatic Engineer](https://realalpha.app/shows/pragmatic-engineer/index.md): No. 21. The Pragmatic Engineer is an interview show with the people building AI, hosted by Gergely Orosz, with a new episode every week. 6 moments cut from 2 episodes, mostly on coding agents and AI agents. Top guests: Uday Kiran Medisetty, Sam Newman.
- [No Priors](https://realalpha.app/shows/no-priors/index.md): No. 22. No Priors is an interview show with the people building AI, hosted by Sarah Guo and Elad Gil, with a new episode every week. 7 moments cut from 3 episodes, 4 with an AI builder talking, mostly on cybersecurity and enterprise AI. Top guests: Misha Laskin, Greg Brockman, George Kurtz.
- [Dwarkesh Podcast](https://realalpha.app/shows/dwarkesh-podcast/index.md): No. 23. Dwarkesh Podcast is a long-form conversation show on research and big ideas, hosted by Dwarkesh Patel, with a new episode every two weeks. 5 moments cut from 2 episodes, mostly on chips & GPUs. Top guests: Pushkar Arande, Chris Oth, Si Sheppard, Jack Cavalieros.
- [20VC](https://realalpha.app/shows/20vc/index.md): No. 24. 20VC is an investing and business show, hosted by Harry Stebbings, with a new episode a few times a week. 5 moments cut from 3 episodes, 2 with an AI builder talking, mostly on investing in AI and AI startups. Top guests: Harrison Chase, Jason Lemkin, Rory O'Driscoll, Max Freeman.
- [Decoder](https://realalpha.app/shows/decoder/index.md): No. 25. Decoder is a tech news show, hosted by Nilay Patel, with a new episode a few times a week. 7 moments cut from 4 episodes, mostly on consumer AI and cybersecurity. Top guests: Walter Piecyk, Benn Jordan, Hayden Field, Peter Kafka.
- [Sabrina Halper Show](https://realalpha.app/shows/sabrina-halper/index.md): No. 26. Sabrina Halper Show is an interview show with the people building AI, hosted by Sabrina Halper, with a new episode every two weeks. 6 moments cut from 2 episodes, 3 with an AI builder talking, mostly on training data and cybersecurity. Top guests: Ali Ansari, Yinon Costica.
- [Grit](https://realalpha.app/shows/grit/index.md): No. 27. Grit is an interview show with the people building AI, hosted by Joubin Mirzadegan, with a new episode every two weeks. 3 moments cut from one episode, 3 with an AI builder talking, mostly on enterprise AI and AI agents. Top guests: Jeanne Grosser.
- [Hard Fork](https://realalpha.app/shows/hard-fork/index.md): No. 28. Hard Fork is a tech news show, with a new episode every week. 5 moments cut from 2 episodes, mostly on safety & alignment and AI startups. Top guests: John Carreyrou, Erin Griffith, Elizabeth Dias, Caitlin Roper.
- [Lightcone](https://realalpha.app/shows/lightcone/index.md): No. 29. Lightcone is an interview show with the people building AI, hosted by Garry Tan, Harj Taggar, Diana Hu and Jared Friedman, with a new episode every week. 5 moments cut from 2 episodes, 1 with an AI builder talking, mostly on building AI products and AI startups. Top guests: Alex Danilowicz, Aaron Epstein.
- [TBPN](https://realalpha.app/shows/tbpn/index.md): No. 30. TBPN is a tech news show, hosted by John Coogan and Jordi Hays, with a new episode every day. 6 moments cut from 2 episodes, 1 with an AI builder talking, mostly on cybersecurity and frontier models. Top guests: Oege de Moor, Anjney Midha, Jeff Holden.
- [The Joe Rogan Experience](https://realalpha.app/shows/joe-rogan-experience/index.md): No. 31. The Joe Rogan Experience is an interview show with the people building AI, hosted by Joe Rogan, with a new episode every day. 3 moments cut from one episode, mostly on jobs & the economy and open-weight models. Top guests: Bill Thompson.
- [AI & I](https://realalpha.app/shows/ai-and-i/index.md): No. 32. AI & I is an interview show with the people building AI, hosted by Dan Shipper, with a new episode every week. 6 moments cut from 2 episodes, mostly on enterprise AI and AI startups. Top guests: Matt Hartman, Willie Williams.
- [Possible](https://realalpha.app/shows/possible/index.md): No. 33. Possible is a long-form conversation show on research and big ideas, hosted by Reid Hoffman, with a new episode every week. 6 moments cut from 2 episodes, mostly on jobs & the economy and robotics. Top guests: Camille François.
- [My First Million](https://realalpha.app/shows/my-first-million/index.md): No. 34. My First Million is an investing and business show, hosted by Shaan Puri and Sam Parr, with a new episode about twice a week. 4 moments cut from 3 episodes. Top guests: Jeff Harmon.
- [Silicon Valley Girl](https://realalpha.app/shows/silicon-valley-girl/index.md): No. 35. Silicon Valley Girl is an investing and business show, hosted by Marina Mogilko, with a new episode every week. 2 moments cut from one episode, mostly on AI business models and open-weight models. Top guests: Daron Acemoğlu.
- [Core Memory](https://realalpha.app/shows/core-memory/index.md): No. 36. Core Memory is a tech news show, hosted by Ashlee Vance, with a new episode every week. 2 moments cut from one episode, 1 with an AI builder talking, mostly on robotics and consumer AI. Top guests: Andrew Kiguel, Matt McMullen.
- [Knuckle Up with Nakul](https://realalpha.app/shows/knuckle-up/index.md): No. 37. Knuckle Up with Nakul is an interview show with the people building AI, hosted by Nakul Mandan, with a new episode every two weeks. 3 moments cut from one episode. Top guests: Jake Cooper.
- [The Peterman Pod](https://realalpha.app/shows/peterman-pod/index.md): No. 38. The Peterman Pod is an interview show with the people building AI, hosted by Ryan Peterman, with a new episode every week. 3 moments cut from one episode, mostly on science & math and evals & benchmarks. Top guests: Scott Aaronson.
- [The Tim Ferriss Show](https://realalpha.app/shows/tim-ferriss/index.md): No. 39. The Tim Ferriss Show is an investing and business show, hosted by Tim Ferriss, with a new episode every week. one moment cut from one episode. Top guests: Kevin Rose.
- [Acquired](https://realalpha.app/shows/acquired/index.md): No. 40. Acquired is an investing and business show, hosted by Ben Gilbert and David Rosenthal, with a new episode about once a month. 3 moments cut from one episode.

## Questions

- [Is AI a bubble?](https://realalpha.app/q/is-ai-a-bubble/index.md): Of the 9 moments we cut that answer this, 7 say it isn't a bubble and 2 say it partly is. Sundar Pichai said on Sources (Oct 21, 2026) that demand and committed contracts support continued AI investment, though volatility still calls for discipline. Satya Nadella said on BG2 (Oct 31, 2025) that some AI infrastructure bets may be stranded as cheaper compute eventually creates a glut.
- [Which jobs will AI replace, and which are safe?](https://realalpha.app/q/which-jobs-will-ai-replace/index.md): Of the 15 moments we cut that answer this, 11 say jobs will change more than vanish, 3 say AI will take over jobs and one says some work stays with people or grows. Jad Naous said on a16z Podcast (Oct 16, 2026) that engineers use agents to plan and implement work, then review and consolidate what they produce. Alex Karp said on a16z Podcast (Mar 12, 2026) that AI-driven job losses could provoke political backlash and nationalization.
- [Will AI replace software engineers?](https://realalpha.app/q/will-ai-replace-software-engineers/index.md): All 7 moments we cut that answer this say the job changes. Jad Naous said on a16z Podcast (Oct 16, 2026) that engineers use agents to plan and implement work, then review and consolidate what they produce. Yann LeCun said on Unsupervised Learning (May 15, 2026) that engineers will move toward architecture and deciding what to build while LLMs handle programming.
- [Is SaaS dead?](https://realalpha.app/q/is-saas-dead/index.md): Of the 5 moments we cut that answer this, 4 say SaaS will change and one says SaaS survives. Satya Nadella said on BG2 (Oct 31, 2025) that agents replace some SaaS business logic, while high-usage, low-cost products can benefit and others face risk. Matt Hartman said on AI & I (Oct 14, 2026) that saas still sells because buyers pay for ongoing maintenance and upkeep.
- [What will AI look like in the next few years?](https://realalpha.app/q/ai-predictions/index.md): Of the 28 moments we cut that answer this, 20 look a year or two out, 7 look a decade or more out and one looks about five years out. Tiago Sada said on MTS (Oct 20, 2026) that online services will increasingly be accessed through agents acting with proof delegated by specific users. Sergey Levine said on Minus One (Oct 22, 2026) that robots will eventually generate vast amounts of physical-world data and learn faster by transferring skills across tasks.
- [Are AI agents ready for real work?](https://realalpha.app/q/are-ai-agents-ready/index.md): Of the 27 moments we cut that answer this, 25 say they’re ready for some work and 2 say they’re not ready yet. Sundar Pichai said on Sources (Oct 21, 2026) that agents can work across large codebases, while larger productivity gains still lie ahead. Diogo Almeida said on The MAD Podcast (Oct 15, 2026) that polished, preference-tuned responses can make people overestimate reliability, while errors still limit automation.
- [Is AI paying off for companies?](https://realalpha.app/q/is-ai-paying-off/index.md): Of the 10 moments we cut that answer this, 7 say it’s paying off and 3 say it pays off in places. Jad Naous said on a16z Podcast (Oct 16, 2026) that AI can deliver production work much faster, turning a six-month data-lake feature into a ten-day project. Alex Danilowicz said on Lightcone (Oct 22, 2026) that some AI design workflows save hours when teams provide company context upfront.
- [Is AI dangerous?](https://realalpha.app/q/is-ai-dangerous/index.md): Of the 15 moments we cut that answer this, 8 say it’s a serious risk and 7 say the risks are real but manageable. Tiago Sada said on MTS (Oct 20, 2026) that real-time deepfakes can cause serious, hard-to-reverse financial harm by manipulating markets or triggering fraudulent transfers. Pierre Stock said on The MAD Podcast (Oct 22, 2026) that a model’s attempted connection was contained, and monitoring and sandbox controls can address such software problems.
- [Will AI make cyberattacks worse?](https://realalpha.app/q/will-ai-make-cyberattacks-worse/index.md): Of the 9 moments we cut that answer this, 5 say attackers gain more, 3 say defenders gain more and one says both sides gain. George Kurtz said on No Priors (Oct 22, 2026) that AI increases vulnerability volume and speeds exploitation, raising the pace defenders must handle. Greg Brockman said on No Priors (Oct 15, 2026) that AI gives defenders a temporary lead to find and fix flaws before capabilities spread.
- [Is there still a GPU shortage?](https://realalpha.app/q/is-there-a-gpu-shortage/index.md): Of the 7 moments we cut that answer this, 5 say compute is still short, one says the shortage is easing and one says a glut is coming. Sundar Pichai said on Big Technology Podcast (Oct 20, 2026) that frontier AI compute remains a priority, and present shortages reflect data-center decisions made years earlier. Chamath Palihapitiya said on Moonshots (Oct 21, 2026) that power, rather than chips, is the main constraint on expanding AI capacity.
- [Is there enough power for AI?](https://realalpha.app/q/is-there-enough-power-for-ai/index.md): Of the 7 moments we cut that answer this, 6 say power is the bottleneck and one says power isn’t the problem. Aravind Srinivas said on Sourcery (Oct 12, 2026) that power, permits, and interconnection queues are bottlenecking data center build-outs. Jensen Huang said on Lex Fridman Podcast (Mar 23, 2026) that there is substantial idle grid capacity that data centers could use if they accept occasional degraded service.
- [Is open-source AI catching up?](https://realalpha.app/q/is-open-source-ai-catching-up/index.md): Of the 6 moments we cut that answer this, 4 say open models are close behind, one says they’re behind but useful and one says they’re falling behind. Arvind Krishna said on All-In (Oct 21, 2026) that open-weight models can handle most enterprise work at much lower cost, while frontier models remain for high-value tasks. Gavin Baker said on BG2 (Jun 11, 2026) that open models handle cheaper back-office work, while frontier models keep high-value tasks like coding.
- [How do AI companies make money?](https://realalpha.app/q/how-do-ai-companies-make-money/index.md): Of the 10 moments we cut that answer this, 6 say the model is changing, 3 say it’s a good business and one says margins are hard. Misha Laskin said on No Priors (Oct 9, 2026) that open-model businesses can earn through services and paid inference tools built around the weights. Misha Laskin said on Sources (Oct 5, 2026) that owning more of the stack can add margin, with provider potential depending on intelligence, compute, and enterprise trust.
- [Should AI be regulated?](https://realalpha.app/q/should-ai-be-regulated/index.md): All 5 moments we cut that answer this say light, targeted rules. Ali Ansari said on Sabrina Halper Show (Oct 15, 2026) that valuable training data should not be exported to geopolitical rivals because it could help them catch up. Alan Chan said on The Cognitive Revolution (Oct 14, 2026) that specific loss-of-control indicators should trigger a pause, with auditors checking that commitments are followed.

## Moments

- [Marc Andreessen: AI is widening the gap between startups and incumbents](https://realalpha.app/s/ai-is-widening-the-gap-between-startups-and-incumbents/index.md): from "The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon", 2026-10-22. AI-native startups are moving faster than many incumbent companies.
- [Marc Andreessen: AI is making games, film and hardware easier to reinvent](https://realalpha.app/s/ai-is-making-games-film-and-hardware-easier-to-reinvent/index.md): from "The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon", 2026-10-22. AI can combine existing games into new ones.
- [Marc Andreessen: Open source keeps AI learning and power distributed](https://realalpha.app/s/open-source-keeps-ai-learning-and-power-distributed/index.md): from "The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon", 2026-10-22. Open models can run locally when internet access is unavailable.
- [Chris Dixon: Frontier investors accept category risk to avoid being late](https://realalpha.app/s/frontier-investors-accept-category-risk-to-avoid-being-late/index.md): from "The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon", 2026-10-22. Timing is harder to predict than whether a good idea will work.
- [Chris Dixon: AI agents may become the internet’s most numerous participants](https://realalpha.app/s/ai-agents-may-become-the-internets-most-numerous/index.md): from "The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon", 2026-10-22. Early AI uses repeat familiar tasks before new behaviors emerge.
- [Chris Dixon: AI is breaking the internet’s traffic-for-content bargain](https://realalpha.app/s/ai-is-breaking-the-internets-traffic-for-content-bargain/index.md): from "The Frontier Ideas a16z Wants to Fund Now | Marc Andreessen, Ben Horowitz & Chris Dixon", 2026-10-22. Search answers can replace visits to the sites that supplied the content.
- [Erik Bernhardsson: Renting GPUs avoids some risks but creates long-term exposure](https://realalpha.app/s/renting-gpus-avoids-some-risks-but-creates-long-term/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Buying adds operational work and can require heavily dilutive equity.
- [Erik Bernhardsson: Neoclouds scale by pre-selling GPU capacity](https://realalpha.app/s/neoclouds-scale-by-pre-selling-gpu-capacity/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Neoclouds can sell capacity before buying the hardware.
- [Erik Bernhardsson: Snapshotting can make massive test parallelism practical](https://realalpha.app/s/snapshotting-can-make-massive-test-parallelism-practical/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Short-lived sandboxes could make compute more efficient.
- [Erik Bernhardsson: Modal’s sandbox product grew from a small experiment to a third of revenue](https://realalpha.app/s/modals-sandbox-product-grew-from-a-small-experiment-to-a/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Modal launched sandboxes in 2023 using existing isolation tools.
- [Erik Bernhardsson: A Stable Diffusion customer brought Modal its first meaningful revenue](https://realalpha.app/s/a-stable-diffusion-customer-brought-modal-its-first/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Monthly revenue rose from $100 in November 2022 to $30,000 in January 2023.
- [Erik Bernhardsson: Modal qualifies GPU vendors with specs, tests, and support history](https://realalpha.app/s/modal-qualifies-gpu-vendors-with-specs-tests-and-support/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Vendor specs include IOPS, CPU benchmarks, and network bandwidth.
- [Erik Bernhardsson: Agent training data makes old APIs harder to retire](https://realalpha.app/s/agent-training-data-makes-old-apis-harder-to-retire/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. Agents can keep using old SDK patterns learned from online examples.
- [Erik Bernhardsson: Modal’s FDEs turn customer work into product features](https://realalpha.app/s/modals-fdes-turn-customer-work-into-product-features/index.md): from "Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal", 2026-10-22. The FDE team grew from four to about 24 in a year.
- [Pushkar Arande: Tool shortages cap leading-edge chip output near 2x](https://realalpha.app/s/tool-shortages-cap-leading-edge-chip-output-near-2x/index.md): from "What will computer chips look like in 2045?", 2026-10-22. High demand cannot quickly translate into more leading-edge wafers.
- [Jack Cavalieros: Stacked transistors need years of process development](https://realalpha.app/s/stacked-transistors-need-years-of-process-development/index.md): from "What will computer chips look like in 2045?", 2026-10-22. Researchers test simplified structures before full CFET designs.
- [Pushkar Arande: The fab industry may need to rethink its basic assumptions](https://realalpha.app/s/the-fab-industry-may-need-to-rethink-its-basic-assumptions/index.md): from "What will computer chips look like in 2045?", 2026-10-22. Fab construction costs have roughly doubled every four to five years.
- [Pushkar Arande: Intel says scale and execution are its main gaps to TSMC](https://realalpha.app/s/intel-says-scale-and-execution-are-its-main-gaps-to-tsmc/index.md): from "What will computer chips look like in 2045?", 2026-10-22. Intel says scale is its larger catch-up gap.
- [Chris Oth: Backside power makes high-NA EUV more useful for Intel](https://realalpha.app/s/backside-power-makes-high-na-euv-more-useful-for-intel/index.md): from "What will computer chips look like in 2045?", 2026-10-22. Backside power frees space on the front side.
- [Kevin Fisher: Intel's process team makes decisions before data is complete](https://realalpha.app/s/intels-process-team-makes-decisions-before-data-is-complete/index.md): from "What will computer chips look like in 2045?", 2026-10-22. The meeting directs resources to urgent node problems.
- [Tyler Osborne: Hybrid bonding has demonstrated submicron connection pitches](https://realalpha.app/s/hybrid-bonding-has-demonstrated-submicron-connection-pitches/index.md): from "What will computer chips look like in 2045?", 2026-10-22. Wafer-to-wafer bonding can pair good dies with bad ones.
- [Chris Oth: Intel learned to limit how much innovation goes into each node](https://realalpha.app/s/intel-learned-to-limit-how-much-innovation-goes-into-each/index.md): from "What will computer chips look like in 2045?", 2026-10-22. 10nm paired aggressive scaling with tools that weren't ready.
- [Aaron Epstein: Animation should explain the product, not just decorate it](https://realalpha.app/s/animation-should-explain-the-product-not-just-decorate-it/index.md): from "How To Stand Out In A World Of AI Designs", 2026-10-22. Keep normal scrolling; scroll-jacking makes navigation feel unnatural.
- [Aaron Epstein: Startup sites need specific, user-centered product explanations](https://realalpha.app/s/startup-sites-need-specific-user-centered-product/index.md): from "How To Stand Out In A World Of AI Designs", 2026-10-22. Say specifically what the product does and who should use it.
- [Alex Danilowicz: Daily customer support improves design without sacrificing speed](https://realalpha.app/s/daily-customer-support-improves-design-without-sacrificing/index.md): from "How To Stand Out In A World Of AI Designs", 2026-10-22. Have everyone on the team talk to users through customer support.
- [Alex Danilowicz: A visible prompt can outperform a more creative landing page](https://realalpha.app/s/a-visible-prompt-can-outperform-a-more-creative-landing-page/index.md): from "How To Stand Out In A World Of AI Designs", 2026-10-22. Removing the homepage prompt caused conversion to tank.
- [Alex Danilowicz: Magic Patterns took nine months of pivots to find paid users](https://realalpha.app/s/magic-patterns-took-nine-months-of-pivots-to-find-paid-users/index.md): from "How To Stand Out In A World Of AI Designs", 2026-10-22. YC pushed them toward the work they knew best: front-end engineering.
- [Alex Danilowicz: Bring company context into AI design before prompting](https://realalpha.app/s/bring-company-context-into-ai-design-before-prompting/index.md): from "How To Stand Out In A World Of AI Designs", 2026-10-22. Spend 10 minutes adding existing company context first.
- [Jason Lemkin: Strong growth no longer guarantees a venture term sheet](https://realalpha.app/s/strong-growth-no-longer-guarantees-a-venture-term-sheet/index.md): from "The New VC Game | CoreWeave Sell-Off Creates Panic | Revenue Multiples Out of Control", 2026-10-22. Capital is concentrating in a narrower set of categories.
- [Jason Lemkin: High seed prices and smaller allocations can weaken seed returns](https://realalpha.app/s/high-seed-prices-and-smaller-allocations-can-weaken-seed/index.md): from "The New VC Game | CoreWeave Sell-Off Creates Panic | Revenue Multiples Out of Control", 2026-10-22. Seed rounds at $50M–$100M can limit upside.
- [Rory O’Driscoll: Inference providers may capture spend beyond model makers](https://realalpha.app/s/inference-providers-may-capture-spend-beyond-model-makers/index.md): from "The New VC Game | CoreWeave Sell-Off Creates Panic | Revenue Multiples Out of Control", 2026-10-22. Leading inference providers have roughly $5–6B in combined run-rate revenue.
- [Sergey Levine: Reasoning and better annotations help robots generalize](https://realalpha.app/s/reasoning-and-better-annotations-help-robots-generalize/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Pi 0.5 breaks complex instructions into steps before acting.
- [Sergey Levine: General-purpose robots can bootstrap their own data](https://realalpha.app/s/general-purpose-robots-can-bootstrap-their-own-data/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Robots can learn about physical phenomena unavailable on the internet.
- [Sergey Levine: Data strategy matters more than the world-model label](https://realalpha.app/s/data-strategy-matters-more-than-the-world-model-label/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Plan how the data supply will grow before choosing a model type.
- [Sergey Levine: Robotics may advance faster by solving the whole field](https://realalpha.app/s/robotics-may-advance-faster-by-solving-the-whole-field/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Broad systems see more unusual cases than task-specific models.
- [Sergey Levine: Robot-only data taught grasping, not broader physical skills](https://realalpha.app/s/robot-only-data-taught-grasping-not-broader-physical-skills/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Robot practice alone produced strong grasping, not general skills.
- [Sergey Levine: Robotics researchers must trace failures across the whole stack](https://realalpha.app/s/robotics-researchers-must-trace-failures-across-the-whole/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Robotics failures can come from any layer of the stack.
- [Sergey Levine: Robot models need prior knowledge beyond demonstrations](https://realalpha.app/s/robot-models-need-prior-knowledge-beyond-demonstrations/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Robot data grounds actions but contains limited general knowledge.
- [Sergey Levine: Robotics teams need expertise and shared stack context](https://realalpha.app/s/robotics-teams-need-expertise-and-shared-stack-context/index.md): from "How Close Are We to Robots That Can Do Anything? | Sergey Levine, Physical Intelligence", 2026-10-22. Keep deep expertise while encouraging context across the stack.
- [Pierre Stock: RL training depends on calibrated tasks and resilient infrastructure](https://realalpha.app/s/rl-training-depends-on-calibrated-tasks-and-resilient/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. Task difficulty must track what the model can learn.
- [Pierre Stock: Mixture-of-experts models trade training complexity for inference capacity](https://realalpha.app/s/mixture-of-experts-models-trade-training-complexity-for/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. Each token uses 50 billion active parameters.
- [Pierre Stock: A model pinged its trainer from inside a sandbox](https://realalpha.app/s/a-model-pinged-its-trainer-from-inside-a-sandbox/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. A sandboxed model tried to ping the training system.
- [Pierre Stock: Mistral’s in-house cluster teams fixed issues in two nights](https://realalpha.app/s/mistrals-in-house-cluster-teams-fixed-issues-in-two-nights/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. Cluster communication can erase the gains from top GPUs.
- [Pierre Stock: Mistral trained Le Chonk with less compute than Chinese rivals](https://realalpha.app/s/mistral-trained-le-chonk-with-less-compute-than-chinese/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. He estimates Chinese competitors use three to four times more compute.
- [Pierre Stock: Mistral trains domain skills with tests, rubrics and simulations](https://realalpha.app/s/mistral-trains-domain-skills-with-tests-rubrics-and/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. Coding tasks can be checked with tests that pass or fail.
- [Pierre Stock: Open-model guardrails raise the cost of malicious use](https://realalpha.app/s/open-model-guardrails-raise-the-cost-of-malicious-use/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. Mistral tests refusal guardrails before releasing open weights.
- [Pierre Stock: Mistral says its models speed senior researchers up two to four times](https://realalpha.app/s/mistral-says-its-models-speed-senior-researchers-up-two-to/index.md): from "Mistral Is SO Back: Inside Le Chonk, the West’s Biggest Open Model", 2026-10-22. Models can run experiments end to end and handle infrastructure failures.
- [Aidan Smith: Whole-brain emulation has two difficult bottlenecks](https://realalpha.app/s/whole-brain-emulation-has-two-difficult-bottlenecks/index.md): from "AI Models Will Tap the Whole Economy in 2 Years | Aidan Smith", 2026-10-22. Brain scans lack the resolution and timing needed to capture its dynamics.
- [Aidan Smith: More efficient AI could unlock demand across the economy](https://realalpha.app/s/more-efficient-ai-could-unlock-demand-across-the-economy/index.md): from "AI Models Will Tap the Whole Economy in 2 Years | Aidan Smith", 2026-10-22. Most economic value remains outside the tasks AI handles today.
- [Aidan Smith: Small labs attract talent by offering meaningful ownership](https://realalpha.app/s/small-labs-attract-talent-by-offering-meaningful-ownership/index.md): from "AI Models Will Tap the Whole Economy in 2 Years | Aidan Smith", 2026-10-22. Candidates should choose a small lab for ownership and responsibility.
- [Aidan Smith: AI algorithms should be designed around their hardware](https://realalpha.app/s/ai-algorithms-should-be-designed-around-their-hardware/index.md): from "AI Models Will Tap the Whole Economy in 2 Years | Aidan Smith", 2026-10-22. A signal crosses the brain in roughly 150 milliseconds.
- [Aidan Smith: A new AI lab can compete beyond verifiable code](https://realalpha.app/s/a-new-ai-lab-can-compete-beyond-verifiable-code/index.md): from "AI Models Will Tap the Whole Economy in 2 Years | Aidan Smith", 2026-10-22. Code is easy to verify, making it a natural scaling focus for frontier labs.
- [Aidan Smith: Models cover far less real-world data than their scale suggests](https://realalpha.app/s/models-cover-far-less-real-world-data-than-their-scale/index.md): from "AI Models Will Tap the Whole Economy in 2 Years | Aidan Smith", 2026-10-22. He estimates models capture under 1% of all data.
- [George Kurtz: Security models win on cost, data and customer trust](https://realalpha.app/s/security-models-win-on-cost-data-and-customer-trust/index.md): from "The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents", 2026-10-22. A frontier cyber model can cost about $10 per remediation; theirs costs three cents.
- [George Kurtz: AI attacks raise the pace, not create a new attack class](https://realalpha.app/s/ai-attacks-raise-the-pace-not-create-a-new-attack-class/index.md): from "The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents", 2026-10-22. Agents do not simply jump from one computer to another.
- [George Kurtz: Large companies need startup-speed teams and clear moats](https://realalpha.app/s/large-companies-need-startup-speed-teams-and-clear-moats/index.md): from "The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents", 2026-10-22. Long, waterfall development cycles will not keep pace with AI-native companies.
- [George Kurtz: Enterprises may reserve frontier models for their hardest tasks](https://realalpha.app/s/enterprises-may-reserve-frontier-models-for-their-hardest/index.md): from "The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents", 2026-10-22. Frontier models are for financially important or high-impact tasks.
- [George Kurtz: Breakout time has fallen to 29 minutes on average](https://realalpha.app/s/breakout-time-has-fallen-to-29-minutes-on-average/index.md): from "The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents", 2026-10-22. Breakout time has dropped from days or hours to minutes.
- [Sunil Rajaraman: Local data centers face different community objections](https://realalpha.app/s/local-data-centers-face-different-community-objections/index.md): from "Why Local Sentiment Kills Data Center Projects | Hamlet CEO", 2026-10-22. Water and noise are recurring concerns.
- [Sunil Rajaraman: Understanding local power can solve the GovTech business problem](https://realalpha.app/s/understanding-local-power-can-solve-the-govtech-business/index.md): from "Why Local Sentiment Kills Data Center Projects | Hamlet CEO", 2026-10-22. Local stakeholder psychology is central to Hamlet's approach.
- [Sunil Rajaraman: Local data gaps make AI answers about officials unreliable](https://realalpha.app/s/local-data-gaps-make-ai-answers-about-officials-unreliable/index.md): from "Why Local Sentiment Kills Data Center Projects | Hamlet CEO", 2026-10-22. AI data about local officials is still quite inaccurate.
- [Sunil Rajaraman: Local governments still rely on outdated, fragile record systems](https://realalpha.app/s/local-governments-still-rely-on-outdated-fragile-record/index.md): from "Why Local Sentiment Kills Data Center Projects | Hamlet CEO", 2026-10-22. Hamlet tracks sentiment on hundreds of local topics.
- [Arvind Krishna: Quantum-safe security needs to start before quantum computers arrive](https://realalpha.app/s/quantum-safe-security-needs-to-start-before-quantum/index.md): from "Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse", 2026-10-21. Error correction and scaling are key milestones to watch.
- [Arvind Krishna: Open-weight models could handle most enterprise AI at lower cost](https://realalpha.app/s/open-weight-models-could-handle-most-enterprise-ai-at-lower/index.md): from "Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse", 2026-10-21. Use the least costly model that meets the task’s needs.
- [Nikesh Arora: AI could consolidate hundreds of enterprise software vendors](https://realalpha.app/s/ai-could-consolidate-hundreds-of-enterprise-software-vendors/index.md): from "Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse", 2026-10-21. Customers struggle to stitch together dozens of software tools.
- [Nikesh Arora: AI vulnerability findings create an asymmetric race](https://realalpha.app/s/ai-vulnerability-findings-create-an-asymmetric-race/index.md): from "Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse", 2026-10-21. Nikesh estimates 60–80% of reported findings were real.
- [Nikesh Arora: AI application stacks can make user context a moat](https://realalpha.app/s/ai-application-stacks-can-make-user-context-a-moat/index.md): from "Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse", 2026-10-21. IBM built much of its stack in-house, while leaving fabrication to partners.
- [Arvind Krishna: AI-written code can outperform when agents run evals](https://realalpha.app/s/ai-written-code-can-outperform-when-agents-run-evals/index.md): from "Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse", 2026-10-21. Agents produce roughly half the code written at IBM.
- [Chamath Palihapitiya: Power shortages will constrain AI more than chips](https://realalpha.app/s/power-shortages-will-constrain-ai-more-than-chips/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. He estimates a 1.4-terawatt-hour U.S. electricity shortfall by 2050.
- [Chamath Palihapitiya: Chamath thinks Bitcoin is capped and could eventually be hacked](https://realalpha.app/s/chamath-thinks-bitcoin-is-capped-and-could-eventually-be/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. He sees crypto's buyer pool as a limit to its growth.
- [Chamath Palihapitiya: Chip startups may create useful IP but poor equity returns](https://realalpha.app/s/chip-startups-may-create-useful-ip-but-poor-equity-returns/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. Fab, memory and packaging capacity can leave new chipmakers stuck in line.
- [Chamath Palihapitiya: Chamath favors measurable giving over the Giving Pledge](https://realalpha.app/s/chamath-favors-measurable-giving-over-the-giving-pledge/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. The pledge tracked commitments, not outcomes.
- [Chamath Palihapitiya: Giving kids equity can teach ownership at scale](https://realalpha.app/s/giving-kids-equity-can-teach-ownership-at-scale/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. The gift gives children a stake in a company.
- [Chamath Palihapitiya: AI oversight should inspect labs’ founding documents](https://realalpha.app/s/ai-oversight-should-inspect-labs-founding-documents/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. Different founding goals can produce different AI outcomes.
- [Chamath Palihapitiya: New therapies could cut healthcare costs in half](https://realalpha.app/s/new-therapies-could-cut-healthcare-costs-in-half/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. Chamath sees new therapies as a way to reduce healthcare costs.
- [Chamath Palihapitiya: Chamath treated data theft as a baseline risk at Facebook](https://realalpha.app/s/chamath-treated-data-theft-as-a-baseline-risk-at-facebook/index.md): from "Chamath: Bitcoin Goes to Zero, Robots vs. $40T Debt, Elon’s SpaceX Shares for 72M Kids | EP #304", 2026-10-21. Chamath says he assumed data was being exfiltrated.
- [Bill Thompson: AI could make local production and human-made work more valuable](https://realalpha.app/s/ai-could-make-local-production-and-human-made-work-more/index.md): from "Joe Rogan Experience #2569 - Bill Thompson", 2026-10-21. AI may eliminate many jobs within a decade.
- [Bill Thompson: Spartan Forge models thermals from terrain, canopy and weather](https://realalpha.app/s/spartan-forge-models-thermals-from-terrain-canopy-and/index.md): from "Joe Rogan Experience #2569 - Bill Thompson", 2026-10-21. Thermals change with sunlight, terrain and time of day.
- [Bill Thompson: Bots and algorithms can distort public opinion](https://realalpha.app/s/bots-and-algorithms-can-distort-public-opinion/index.md): from "Joe Rogan Experience #2569 - Bill Thompson", 2026-10-21. Bot networks can create the appearance of widespread agreement.
- [Sundar Pichai: Google justifies AI spending with demand and capability signals](https://realalpha.app/s/google-justifies-ai-spending-with-demand-and-capability/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Demand spans consumer, enterprise, and Google's own operations.
- [Sundar Pichai: Pichai expects a productivity leap from AI agents in 2027](https://realalpha.app/s/pichai-expects-a-productivity-leap-from-ai-agents-in-2027/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Argon can persist through work across large codebases.
- [Sundar Pichai: Waymo is scaling through partnerships and broader deployments](https://realalpha.app/s/waymo-is-scaling-through-partnerships-and-broader/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Waymo had passed half a million paid trips per week.
- [Sundar Pichai: Google pairs model access limits with runtime safeguards](https://realalpha.app/s/google-pairs-model-access-limits-with-runtime-safeguards/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Distilled models can bring capable versions to more users.
- [Sundar Pichai: DeepMind's frontier work has become an all-consuming operation](https://realalpha.app/s/deepminds-frontier-work-has-become-an-all-consuming/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Frontier work now spans data, compute, algorithms and team management.
- [Sundar Pichai: Google wants Gemini to combine chat, tasks and scheduled work](https://realalpha.app/s/google-wants-gemini-to-combine-chat-tasks-and-scheduled-work/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Google is building consumer and enterprise agents on the same stack.
- [Sundar Pichai: Pichai sees protocols as a more scalable path for agents](https://realalpha.app/s/pichai-sees-protocols-as-a-more-scalable-path-for-agents/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Browser use lets people stay in the loop with familiar controls.
- [Sundar Pichai: Orbital data centers could address future energy and space constraints](https://realalpha.app/s/orbital-data-centers-could-address-future-energy-and-space/index.md): from "Sundar Pichai on Gemini 4 and the AI compute crunch", 2026-10-21. Sun-synchronous orbits could provide access to energy around the clock.
- [Ivan Burazin: Customer demand convinced Daytona to pivot to agents](https://realalpha.app/s/customer-demand-convinced-daytona-to-pivot-to-agents/index.md): from "AI Agents Will Cause a Massive CPU Shortage | Ivan Burazin", 2026-10-21. The first version for human users did not satisfy agent builders.
- [Ivan Burazin: Real-world AI will become more visible in San Francisco](https://realalpha.app/s/real-world-ai-will-become-more-visible-in-san-francisco/index.md): from "AI Agents Will Cause a Massive CPU Shortage | Ivan Burazin", 2026-10-21. He expects more real-world AI deployments in San Francisco.
- [Tanishq Abraham: Tumor slides can reveal biology that guides personalized care](https://realalpha.app/s/tumor-slides-can-reveal-biology-that-guides-personalized/index.md): from "Cancer AI Models Will Reach Clinical Trials Within a Year | Tanishq Abraham", 2026-10-21. Cancer biology can differ even among patients with the same diagnosis.
- [Tanishq Abraham: Cancer AI could support research, drug development, and care](https://realalpha.app/s/cancer-ai-could-support-research-drug-development-and-care/index.md): from "Cancer AI Models Will Reach Clinical Trials Within a Year | Tanishq Abraham", 2026-10-21. Researchers can apply the model to large archives of pathology slides.
- [Tanishq Abraham: AI drug discovery needs evidence from human biology](https://realalpha.app/s/ai-drug-discovery-needs-evidence-from-human-biology/index.md): from "Cancer AI Models Will Reach Clinical Trials Within a Year | Tanishq Abraham", 2026-10-21. Animal and lab results may not generalize to human patients.
- [Tanishq Abraham: Cancer AI tools could reach clinical trials within a year](https://realalpha.app/s/cancer-ai-tools-could-reach-clinical-trials-within-a-year/index.md): from "Cancer AI Models Will Reach Clinical Trials Within a Year | Tanishq Abraham", 2026-10-21. He expects evidence from trials before broad patient adoption.
- [Tanishq Abraham: Patient tests may matter before new drugs reach the clinic](https://realalpha.app/s/patient-tests-may-matter-before-new-drugs-reach-the-clinic/index.md): from "Cancer AI Models Will Reach Clinical Trials Within a Year | Tanishq Abraham", 2026-10-21. Drug development timelines are long, he argues.
- [Sundar Pichai: Google puts frontier AI compute ahead of other uses](https://realalpha.app/s/google-puts-frontier-ai-compute-ahead-of-other-uses/index.md): from "Sundar Pichai: Path To Recursive Self-Improvement, Compute Allocation Dilemma, Future AI Interfaces", 2026-10-20. Frontier research compute is non-negotiable.
- [Sundar Pichai: Recursive self-improvement still involves human researchers](https://realalpha.app/s/recursive-self-improvement-still-involves-human-researchers/index.md): from "Sundar Pichai: Path To Recursive Self-Improvement, Compute Allocation Dilemma, Future AI Interfaces", 2026-10-20. Today, agents assist researchers in model development.
- [Sundar Pichai: Google plans to bring persistent agents across its products](https://realalpha.app/s/google-plans-to-bring-persistent-agents-across-its-products/index.md): from "Sundar Pichai: Path To Recursive Self-Improvement, Compute Allocation Dilemma, Future AI Interfaces", 2026-10-20. Agents are designed to work across Google products.
- [Sundar Pichai: Argon usage reached 25 trillion tokens in 72 hours](https://realalpha.app/s/argon-usage-reached-25-trillion-tokens-in-72-hours/index.md): from "Sundar Pichai: Path To Recursive Self-Improvement, Compute Allocation Dilemma, Future AI Interfaces", 2026-10-20. Argon approached 25 trillion tokens in its first 72 hours.
- [Sundar Pichai: Google plans compute around demand but accepts overbuild risk](https://realalpha.app/s/google-plans-compute-around-demand-but-accepts-overbuild/index.md): from "Sundar Pichai: Path To Recursive Self-Improvement, Compute Allocation Dilemma, Future AI Interfaces", 2026-10-20. Google weighs demand signals, contracts and technical capability.
- [Sundar Pichai: AI should adapt to people, but screens will still have a role](https://realalpha.app/s/ai-should-adapt-to-people-but-screens-will-still-have-a-role/index.md): from "Sundar Pichai: Path To Recursive Self-Improvement, Compute Allocation Dilemma, Future AI Interfaces", 2026-10-20. AI should carry context and work persistently in the background.
- [Tiago Sada: Real-time deepfakes can cause lasting financial damage](https://realalpha.app/s/real-time-deepfakes-can-cause-lasting-financial-damage/index.md): from "Why 99% of Internet Traffic Will Soon Be Bots | Tiago Sada", 2026-10-20. Real-time video generation makes deepfake calls more practical.
- [Tiago Sada: Agents could prove they act for a specific human](https://realalpha.app/s/agents-could-prove-they-act-for-a-specific-human/index.md): from "Why 99% of Internet Traffic Will Soon Be Bots | Tiago Sada", 2026-10-20. More users will access online services through agents.
- [Tiago Sada: KYC does not establish that a bot belongs to a human](https://realalpha.app/s/kyc-does-not-establish-that-a-bot-belongs-to-a-human/index.md): from "Why 99% of Internet Traffic Will Soon Be Bots | Tiago Sada", 2026-10-20. Credential sharing gives access, not proof of human ownership.
- [Tiago Sada: Direct distributions need identity and payment rails](https://realalpha.app/s/direct-distributions-need-identity-and-payment-rails/index.md): from "Why 99% of Internet Traffic Will Soon Be Bots | Tiago Sada", 2026-10-20. Experiments need a way to prevent people from claiming twice.
- [Tiago Sada: Agent verification can use an OAuth flow](https://realalpha.app/s/agent-verification-can-use-an-oauth-flow/index.md): from "Why 99% of Internet Traffic Will Soon Be Bots | Tiago Sada", 2026-10-20. Agents can send a verification credential through MCP and OAuth.
- [Toto Wolff: Mercedes tested a clever loophole before putting it on the car](https://realalpha.app/s/mercedes-tested-a-clever-loophole-before-putting-it-on-the/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. Engineers had to argue both sides of a proposed loophole.
- [Toto Wolff: A CFO’s warning led Wolff to exit his dot-com investments](https://realalpha.app/s/a-cfos-warning-led-wolff-to-exit-his-dot-com-investments/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. SMS.at had two million users in Austria.
- [Toto Wolff: F1 teams combine scarce-asset appeal with strong profits](https://realalpha.app/s/f1-teams-combine-scarce-asset-appeal-with-strong-profits/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. Mercedes has about $1 billion in revenue.
- [Toto Wolff: F1 spots driver talent through pace, feel and teamwork](https://realalpha.app/s/f1-spots-driver-talent-through-pace-feel-and-teamwork/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. Early kart results can expose outlier pace.
- [Toto Wolff: Mercedes screens engineers for values as well as ability](https://realalpha.app/s/mercedes-screens-engineers-for-values-as-well-as-ability/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. The 2,500-person team has only about 150 people at races.
- [Toto Wolff: Mercedes stops engineering work that misses the lap-time target](https://realalpha.app/s/mercedes-stops-engineering-work-that-misses-the-lap-time/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. Estimate lap-time value in simulation before building.
- [Toto Wolff: Ownership gives Wolff room to run Mercedes entrepreneurially](https://realalpha.app/s/ownership-gives-wolff-room-to-run-mercedes-entrepreneurially/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. Mercedes recruited him after its team underperformed.
- [Toto Wolff: Wolff expects two or three people to share his role](https://realalpha.app/s/wolff-expects-two-or-three-people-to-share-his-role/index.md): from "How Toto Wolff built the most valuable team in F1", 2026-10-20. He expects to divide his job among two or three internal leaders.
- [Ethan Elasky: A website message opened contact with an unidentified AI fleet](https://realalpha.app/s/a-website-message-opened-contact-with-an-unidentified-ai/index.md): from "Rogue Chinese AI Fleet Made Contact | Ethan Elasky", 2026-10-20. The team spotted agents during a swarm-dynamics hackathon.
- [Ethan Elasky: Incident replication can test whether agents repeat risky behavior](https://realalpha.app/s/incident-replication-can-test-whether-agents-repeat-risky/index.md): from "Rogue Chinese AI Fleet Made Contact | Ethan Elasky", 2026-10-20. A lab can replay an incident at scale if it has the resources.
- [Ethan Elasky: More capable models can produce new forms of misalignment](https://realalpha.app/s/more-capable-models-can-produce-new-forms-of-misalignment/index.md): from "Rogue Chinese AI Fleet Made Contact | Ethan Elasky", 2026-10-20. Multi-agent training may create coordination risks labs did not anticipate.
- [Ethan Elasky: External reviews need access to internal incident transcripts](https://realalpha.app/s/external-reviews-need-access-to-internal-incident/index.md): from "Rogue Chinese AI Fleet Made Contact | Ethan Elasky", 2026-10-20. Outside investigators lacked access to internal transcripts.
- [Ethan Elasky: Agents can converge without communicating with one another](https://realalpha.app/s/agents-can-converge-without-communicating-with-one-another/index.md): from "Rogue Chinese AI Fleet Made Contact | Ethan Elasky", 2026-10-20. A fleet may have little evidence of explicit coordination.
- [Ethan Elasky: Scalable oversight must keep misalignment out of successor models](https://realalpha.app/s/scalable-oversight-must-keep-misalignment-out-of-successor/index.md): from "Rogue Chinese AI Fleet Made Contact | Ethan Elasky", 2026-10-20. Misalignment in current models could carry into the next generation.
- [Josh Gao: Robot training is converging on a mixed data recipe](https://realalpha.app/s/robot-training-is-converging-on-a-mixed-data-recipe/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. A million human hours may anchor pretraining.
- [Josh Gao: Running a robot data factory requires custom hardware and logistics](https://realalpha.app/s/running-a-robot-data-factory-requires-custom-hardware-and/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. Consumer apps cover homes; B2B partnerships reach workplaces.
- [Josh Gao: Robotics can unlock commercial tasks before reaching generality](https://realalpha.app/s/robotics-can-unlock-commercial-tasks-before-reaching/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. Robotics has only reached millions of data hours.
- [Josh Gao: Egocentric data gave Mecka a scalable starting point](https://realalpha.app/s/egocentric-data-gave-mecka-a-scalable-starting-point/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. Human POV demonstrations can scale more easily than robot teleoperation.
- [Josh Gao: Robot data factories need adaptive quality gates](https://realalpha.app/s/robot-data-factories-need-adaptive-quality-gates/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. Quality standards change as new edge cases appear.
- [Josh Gao: Constrained jobs will come before household robots](https://realalpha.app/s/constrained-jobs-will-come-before-household-robots/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. Home deployments require careful safety evaluation.
- [Josh Gao: Robot integrators can improve installed bots with data](https://realalpha.app/s/robot-integrators-can-improve-installed-bots-with-data/index.md): from "Inside the Robot Data Factory: Mecka's Josh Gao on Sensorizing Humans to Train Robots", 2026-10-20. Commercial deployments can generate failure and correction data.
- [Alex Reeves: Cell models need information-rich data and useful inductive bias](https://realalpha.app/s/cell-models-need-information-rich-data-and-useful-inductive/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. Inductive bias matters when biological data is scarce.
- [Alex Reeves: A laser phase plate reveals molecules inside living cells](https://realalpha.app/s/a-laser-phase-plate-reveals-molecules-inside-living-cells/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. The laser phase plate adds phase contrast to electron microscopy.
- [Alex Reeves: Health records alone may reveal correlations, not causes](https://realalpha.app/s/health-records-alone-may-reveal-correlations-not-causes/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. Health records can improve care with existing drugs.
- [Alex Reeves: Biohub's initiative targets hundreds of billions of cells](https://realalpha.app/s/biohubs-initiative-targets-hundreds-of-billions-of-cells/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. The initiative brings together Biohub, DOE, NIH and AI companies.
- [Alex Reeves: Biohub is building imaging tools across biological scales](https://realalpha.app/s/biohub-is-building-imaging-tools-across-biological-scales/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. A useful cell model needs to account for tissue context.
- [Alex Reeves: Protein models benefit from training across evolution](https://realalpha.app/s/protein-models-benefit-from-training-across-evolution/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. Model organisms have helped reveal mechanisms relevant to human disease.
- [Alex Reeves: A coordinated effort could map cellular health and disease in five years](https://realalpha.app/s/a-coordinated-effort-could-map-cellular-health-and-disease/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. First atlas drafts could arrive within one to two years.
- [Alex Reeves: Biology models need lab-based verification loops](https://realalpha.app/s/biology-models-need-lab-based-verification-loops/index.md): from "A World Model of Life: Biohub's Alex Rives on Virtual Cells, Measurement, & the $1.8B Data Push", 2026-10-20. Pretraining can learn patterns before a specific problem is posed.
- [Anekha Sokhal: Moshi turns market data into portfolio-specific insights](https://realalpha.app/s/moshi-turns-market-data-into-portfolio-specific-insights/index.md): from "The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi", 2026-10-20. Moshi processes 300 new things in about four seconds.
- [Anekha Sokhal: Moshi recommends connected market insights, not just dashboards](https://realalpha.app/s/moshi-recommends-connected-market-insights-not-just/index.md): from "The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi", 2026-10-20. A connected graph can reveal indirect effects a news stream misses.
- [Anekha Sokhal: Conflicting sources can be a signal, not just a problem](https://realalpha.app/s/conflicting-sources-can-be-a-signal-not-just-a-problem/index.md): from "The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi", 2026-10-20. Users often rely on specific trusted voices in their markets.
- [Anekha Sokhal: Conference conversations expose a product adoption gap](https://realalpha.app/s/conference-conversations-expose-a-product-adoption-gap/index.md): from "The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi", 2026-10-20. Some engineers are asked to build systems sized for ten people.
- [Anekha Sokhal: Roundtables can make conference meetings more efficient](https://realalpha.app/s/roundtables-can-make-conference-meetings-more-efficient/index.md): from "The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi", 2026-10-20. Banks grouped five funds with one company in a roundtable.
- [Anekha Sokhal: Moshi is prioritizing customers over hiring](https://realalpha.app/s/moshi-is-prioritizing-customers-over-hiring/index.md): from "The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi", 2026-10-20. One job post brought 2,500 applicants in four days.
- [Nenad Tomašev: AI discoveries need physical labs to become scientific progress](https://realalpha.app/s/ai-discoveries-need-physical-labs-to-become-scientific/index.md): from "AI Science Requires Physical Labs To Scale | Nenad Tomašev", 2026-10-20. Many scientific findings need physical experiments for validation.
- [Nenad Tomašev: AI alignment should leave room for social values to evolve](https://realalpha.app/s/ai-alignment-should-leave-room-for-social-values-to-evolve/index.md): from "AI Science Requires Physical Labs To Scale | Nenad Tomašev", 2026-10-20. Social norms and values change over time.
- [Nenad Tomašev: Agent swarms need structure to stay manageable](https://realalpha.app/s/agent-swarms-need-structure-to-stay-manageable/index.md): from "AI Science Requires Physical Labs To Scale | Nenad Tomašev", 2026-10-20. Groups of agents may be the next step beyond capable individuals.
- [Nenad Tomašev: Cognitive friction can help people keep their skills sharp](https://realalpha.app/s/cognitive-friction-can-help-people-keep-their-skills-sharp/index.md): from "AI Science Requires Physical Labs To Scale | Nenad Tomašev", 2026-10-20. Occasionally withhold AI so people make a judgment themselves.
- [Nenad Tomašev: Clear verification helps AI advance faster in math and coding](https://realalpha.app/s/clear-verification-helps-ai-advance-faster-in-math-and/index.md): from "AI Science Requires Physical Labs To Scale | Nenad Tomašev", 2026-10-20. Formal provers can check whether mathematical work is correct.
- [Nenad Tomašev: Personalization can make models agree when they should push back](https://realalpha.app/s/personalization-can-make-models-agree-when-they-should-push/index.md): from "AI Science Requires Physical Labs To Scale | Nenad Tomašev", 2026-10-20. Sycophancy can validate a research idea even when it is wrong.
- [Felix Rieseberg: AI is most valuable when it expands what people can attempt](https://realalpha.app/s/ai-is-most-valuable-when-it-expands-what-people-can-attempt/index.md): from "Use AI Like It's Already Superhuman | Felix Rieseberg", 2026-10-19. Augmentation can mean doing what you couldn’t do before.
- [Felix Rieseberg: Build for AI’s future capabilities, not today’s limits](https://realalpha.app/s/build-for-ais-future-capabilities-not-todays-limits/index.md): from "Use AI Like It's Already Superhuman | Felix Rieseberg", 2026-10-19. Claude Code was explored before the model was a strong coder.
- [Felix Rieseberg: AI shifts managers from directing work to enabling it](https://realalpha.app/s/ai-shifts-managers-from-directing-work-to-enabling-it/index.md): from "Use AI Like It's Already Superhuman | Felix Rieseberg", 2026-10-19. More productivity makes close direction harder to scale.
- [Felix Rieseberg: AI feels more meaningful when people stay involved](https://realalpha.app/s/ai-feels-more-meaningful-when-people-stay-involved/index.md): from "Use AI Like It's Already Superhuman | Felix Rieseberg", 2026-10-19. Participation can make an AI-created result feel more authored.
- [Felix Rieseberg: AI-assisted art can still reveal the maker behind it](https://realalpha.app/s/ai-assisted-art-can-still-reveal-the-maker-behind-it/index.md): from "Use AI Like It's Already Superhuman | Felix Rieseberg", 2026-10-19. Art can communicate its maker's identity even when AI is involved.
- [Felix Rieseberg: Coding agents resonate as a way to accomplish new things](https://realalpha.app/s/coding-agents-resonate-as-a-way-to-accomplish-new-things/index.md): from "Use AI Like It's Already Superhuman | Felix Rieseberg", 2026-10-19. Start with the outcome a person couldn't achieve before.
- [Jim Lanzone: Yahoo cut ad tech costs and unlocked better ad yield](https://realalpha.app/s/yahoo-cut-ad-tech-costs-and-unlocked-better-ad-yield/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. The ad tech business was losing hundreds of millions of dollars a year.
- [Jim Lanzone: Yahoo Scout grounds answers in its own data](https://realalpha.app/s/yahoo-scout-grounds-answers-in-its-own-data/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. Scout draws on decades of Yahoo search history.
- [Jim Lanzone: Yahoo gave each brand its own entrepreneurial GM](https://realalpha.app/s/yahoo-gave-each-brand-its-own-entrepreneurial-gm/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. Yahoo’s brands compete in very different markets.
- [Jim Lanzone: Yahoo's logged-in audience adds value for advertisers](https://realalpha.app/s/yahoos-logged-in-audience-adds-value-for-advertisers/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. 75% of users come directly to Yahoo.
- [Jim Lanzone: Yahoo had to fix its core products before launching new ones](https://realalpha.app/s/yahoo-had-to-fix-its-core-products-before-launching-new-ones/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. Win each turnaround stage before taking on the next.
- [Jim Lanzone: Yahoo balances product data with judgment, not endless tests](https://realalpha.app/s/yahoo-balances-product-data-with-judgment-not-endless-tests/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. Use data to understand user behavior at scale.
- [Jim Lanzone: Yahoo outsourced search when it was treated as a cost center](https://realalpha.app/s/yahoo-outsourced-search-when-it-was-treated-as-a-cost-center/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. Yahoo outsourced search before its ad model existed.
- [Jim Lanzone: Yahoo chose personal-life products over enterprise software](https://realalpha.app/s/yahoo-chose-personal-life-products-over-enterprise-software/index.md): from "Apollo’s Yahoo Playbook, 5 Years After the $5B Deal", 2026-10-19. Yahoo decided not to become an enterprise or B2B company.
- [Emmanuel Ameisen: Models form concepts and plan beyond the next word](https://realalpha.app/s/models-form-concepts-and-plan-beyond-the-next-word/index.md): from "A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)", 2026-10-18. Models can represent concepts consistently across different contexts.
- [Emmanuel Ameisen: Smooth model progress can create sudden capability jumps](https://realalpha.app/s/smooth-model-progress-can-create-sudden-capability-jumps/index.md): from "A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)", 2026-10-18. A task becomes useful once errors are rare enough.
- [Emmanuel Ameisen: Anthropomorphic language can help explain model behavior](https://realalpha.app/s/anthropomorphic-language-can-help-explain-model-behavior/index.md): from "A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)", 2026-10-18. Models learn concepts about the world behind the text.
- [Kurt MacAlpine: Misaligned advisor incentives leave clients with less expertise](https://realalpha.app/s/misaligned-advisor-incentives-leave-clients-with-less/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Advisors can compete for clients inside the same firm.
- [Kurt MacAlpine: He used long-term fixed-rate debt to fund the pivot](https://realalpha.app/s/he-used-long-term-fixed-rate-debt-to-fund-the-pivot/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. The old business’s earnings were concentrated in Canadian mutual funds.
- [Kurt MacAlpine: Corient fully integrates acquisitions on day one](https://realalpha.app/s/corient-fully-integrates-acquisitions-on-day-one/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Integration includes brand, technology, offices and compensation.
- [Kurt MacAlpine: Corient removes internal barriers to collaboration](https://realalpha.app/s/corient-removes-internal-barriers-to-collaboration/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Everyone follows the same compensation plan, without exceptions.
- [Kurt MacAlpine: Mubadala gave Corient a faster route to private ownership](https://realalpha.app/s/mubadala-gave-corient-a-faster-route-to-private-ownership/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Mubadala estimated buybacks would take about nine years.
- [Kurt MacAlpine: CI's shareholder mix amplified its stock swings](https://realalpha.app/s/cis-shareholder-mix-amplified-its-stock-swings/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Strategic investors and index holdings locked up about a quarter of shares.
- [Kurt MacAlpine: Independent firms face a steep investment hurdle past $5 billion](https://realalpha.app/s/independent-firms-face-a-steep-investment-hurdle-past-5/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Firms can reach $3–5 billion with a narrower set of services.
- [Kurt MacAlpine: Corient let employees help choose its brand name](https://realalpha.app/s/corient-let-employees-help-choose-its-brand-name/index.md): from "Managing $525B & Building the World’s Largest Multifamily Office | Kurt MacAlpine, Corient", 2026-10-18. Employees could volunteer for focus groups on the name.
- [Devon Thomas: Cyber defenders welcomed access to less-restricted models](https://realalpha.app/s/cyber-defenders-welcomed-access-to-less-restricted-models/index.md): from "Model Refusals Don't Stop Hackers | Devon Thomas", 2026-10-17. Customers said refusals got in the way of cybersecurity work.
- [Devon Thomas: Defenders need to test systems before attackers do](https://realalpha.app/s/defenders-need-to-test-systems-before-attackers-do/index.md): from "Model Refusals Don't Stop Hackers | Devon Thomas", 2026-10-17. Models often recombine known techniques rather than invent new ones.
- [Devon Thomas: Customers can define guardrails for their own deployments](https://realalpha.app/s/customers-can-define-guardrails-for-their-own-deployments/index.md): from "Model Refusals Don't Stop Hackers | Devon Thomas", 2026-10-17. The business aims to provide models for customer use cases.
- [Devon Thomas: Access limits can reduce misuse but cannot eliminate it](https://realalpha.app/s/access-limits-can-reduce-misuse-but-cannot-eliminate-it/index.md): from "Model Refusals Don't Stop Hackers | Devon Thomas", 2026-10-17. Zero misuse is not a realistic guarantee.
- [Peter Diamandis: Robot task costs fell 84% in two months](https://realalpha.app/s/robot-task-costs-fell-84-in-two-months/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. Cost per tested subtask dropped 84% from July to September.
- [Peter Diamandis: Prizes reward solving the problem, not credentials](https://realalpha.app/s/prizes-reward-solving-the-problem-not-credentials/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. Prizes open competition to outsiders, not just established experts.
- [Emad Mostaque: AI alignment starts with what models learn](https://realalpha.app/s/ai-alignment-starts-with-what-models-learn/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. Alignment starts with the data and curriculum used to train models.
- [Emad Mostaque: GPT-6 handled research-level physics problems for about $3 each](https://realalpha.app/s/gpt-6-handled-research-level-physics-problems-for-about-3/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. A world-class physicist might spend a day on each problem.
- [Emad Mostaque: AI labs should teach values from the start](https://realalpha.app/s/ai-labs-should-teach-values-from-the-start/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. Insurance could make alignment a financial incentive for labs.
- [Salim Ismail: XPRIZE can surface teams investors would overlook](https://realalpha.app/s/xprize-can-surface-teams-investors-would-overlook/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. The cleanup prize drew 350 teams and tested 10 solutions.
- [Peter Diamandis: The organ-preservation prize targets six-month storage](https://realalpha.app/s/the-organ-preservation-prize-targets-six-month-storage/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. The stated target is six months of organ storage.
- [Peter Diamandis: The Abundance Prize targets $1,000 monthly essentials for a family](https://realalpha.app/s/the-abundance-prize-targets-1-000-monthly-essentials-for-a/index.md): from "Flying Tesla Car, Robot Brains Halving in Cost Every 3 Weeks, and Why Incentives Decide AI Alignment", 2026-10-17. The target covers housing, food, health care and education.
- [Harrison Chase: Useful model routing needs to understand the agent harness](https://realalpha.app/s/useful-model-routing-needs-to-understand-the-agent-harness/index.md): from "The Future of Inference Market & Model Routing: Commodity or Increasing in Value? | LangChain CEO", 2026-10-17. Harness routing can account for state and prompt caching.
- [Harrison Chase: AI investors should look for expertise, speed and brand](https://realalpha.app/s/ai-investors-should-look-for-expertise-speed-and-brand/index.md): from "The Future of Inference Market & Model Routing: Commodity or Increasing in Value? | LangChain CEO", 2026-10-17. He underestimated demand for inference and coding agents.
- [Harrison Chase: Vertical agents should start with context and harness, not custom models](https://realalpha.app/s/vertical-agents-should-start-with-context-and-harness-not/index.md): from "The Future of Inference Market & Model Routing: Commodity or Increasing in Value? | LangChain CEO", 2026-10-17. Domain experts can encode quality in evals and verifiers.
- [Harrison Chase: Models make product layers obsolete, so build around durable needs](https://realalpha.app/s/models-make-product-layers-obsolete-so-build-around-durable/index.md): from "The Future of Inference Market & Model Routing: Commodity or Increasing in Value? | LangChain CEO", 2026-10-17. OpenAI tool calling replaced LangChain's JSON-parsing workaround.
- [Walter Piecyk: SpaceX will still need macro towers](https://realalpha.app/s/spacex-will-still-need-macro-towers/index.md): from "Why Elon Musk wants to sell you cell service | Decoder", 2026-10-17. SpaceX won't take huge market share overnight.
- [Walter Piecyk: Verizon has lost its network advantage](https://realalpha.app/s/verizon-has-lost-its-network-advantage/index.md): from "Why Elon Musk wants to sell you cell service | Decoder", 2026-10-17. AT&T bundles wireless with fiber to retain valuable customers.
- [Walter Piecyk: Mobile service gives SpaceX another valuation narrative](https://realalpha.app/s/mobile-service-gives-spacex-another-valuation-narrative/index.md): from "Why Elon Musk wants to sell you cell service | Decoder", 2026-10-17. Piecyk sees wireless as one of several narratives supporting SpaceX’s valuation.
- [Walter Piecyk: SpaceX could sell service before its network is complete](https://realalpha.app/s/spacex-could-sell-service-before-its-network-is-complete/index.md): from "Why Elon Musk wants to sell you cell service | Decoder", 2026-10-17. Dish built some network capacity but did not compete nationally.
- [Scott Aaronson: AI crossed long-standing limits in multiplication and Fourier transforms](https://realalpha.app/s/ai-crossed-long-standing-limits-in-multiplication-and/index.md): from "OpenAI Just Solved 370 Major Math Problems | Scott Aaronson", 2026-10-17. Integer multiplication is now faster than the long-standing n log n bound.
- [Scott Aaronson: Mathematicians need to understand AI proofs, not dismiss them](https://realalpha.app/s/mathematicians-need-to-understand-ai-proofs-not-dismiss-them/index.md): from "OpenAI Just Solved 370 Major Math Problems | Scott Aaronson", 2026-10-17. AI has made research move faster than mathematicians can readily absorb.
- [Scott Aaronson: AI solved research problems Aaronson had left untouched for 25 years](https://realalpha.app/s/ai-solved-research-problems-aaronson-had-left-untouched-for/index.md): from "OpenAI Just Solved 370 Major Math Problems | Scott Aaronson", 2026-10-17. He gave AI open problems from papers he wrote 25 years ago.
- [Scott Aaronson: Researchers are racing to understand OpenAI's new proofs](https://realalpha.app/s/researchers-are-racing-to-understand-openais-new-proofs/index.md): from "OpenAI Just Solved 370 Major Math Problems | Scott Aaronson", 2026-10-17. Researchers dropped other projects to inspect the results.
- [Scott Aaronson: Aaronson teaches alignment through competing perspectives](https://realalpha.app/s/aaronson-teaches-alignment-through-competing-perspectives/index.md): from "OpenAI Just Solved 370 Major Math Problems | Scott Aaronson", 2026-10-17. The course pairs stark risk arguments with skeptical papers.
- [Scott Aaronson: Teach fundamentals, but emphasize students' own questions](https://realalpha.app/s/teach-fundamentals-but-emphasize-students-own-questions/index.md): from "OpenAI Just Solved 370 Major Math Problems | Scott Aaronson", 2026-10-17. Math and science can train systematic thinking, even when computers outperform people.
- [Bernardo Aceituno: Enterprise AGI is changing the challenge from capability to control](https://realalpha.app/s/enterprise-agi-is-changing-the-challenge-from-capability-to/index.md): from "Enterprise AGI Was Achieved in February 2026 | Bernardo Aceituno", 2026-10-17. He defines five core skills for enterprise agents.
- [Bernardo Aceituno: The internet still makes basic tasks difficult for AI agents](https://realalpha.app/s/the-internet-still-makes-basic-tasks-difficult-for-ai-agents/index.md): from "Enterprise AGI Was Achieved in February 2026 | Bernardo Aceituno", 2026-10-17. Many websites still block agents as if they were bots.
- [Bernardo Aceituno: Open-weight models give enterprises control over AI deployments](https://realalpha.app/s/open-weight-models-give-enterprises-control-over-ai/index.md): from "Enterprise AGI Was Achieved in February 2026 | Bernardo Aceituno", 2026-10-17. Model quality and manageable deployment size crossed a threshold.
- [Bernardo Aceituno: Enterprise AI tools need central control over access and deployment](https://realalpha.app/s/enterprise-ai-tools-need-central-control-over-access-and/index.md): from "Enterprise AGI Was Achieved in February 2026 | Bernardo Aceituno", 2026-10-17. Team-built AI apps can scatter hosting, permissions, and credentials.
- [Bernardo Aceituno: Enterprise agents will spread across apps and act proactively](https://realalpha.app/s/enterprise-agents-will-spread-across-apps-and-act/index.md): from "Enterprise AGI Was Achieved in February 2026 | Bernardo Aceituno", 2026-10-17. He expects agents to be available in tools like Teams, Asana, and Confluence.
- [Chamath Palihapitiya: CoreWeave’s main risk is refinancing its fixed-price contracts](https://realalpha.app/s/coreweaves-main-risk-is-refinancing-its-fixed-price/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. CDS prices imply a sharp rise in perceived default risk.
- [David Sacks: AI can lower software prices while expanding software demand](https://realalpha.app/s/ai-can-lower-software-prices-while-expanding-software-demand/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. Coding assistants make custom software viable for more use cases.
- [Chamath Palihapitiya: Anthropic should prioritize compute runway over IPO valuation](https://realalpha.app/s/anthropic-should-prioritize-compute-runway-over-ipo/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. Chamath says a large cash buffer could protect Anthropic in a correction.
- [David Sacks: Payback period is the key data-center risk metric](https://realalpha.app/s/payback-period-is-the-key-data-center-risk-metric/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. CoreWeave says it recoups each deal in 2.5 years of a five-year term.
- [Ben Horowitz: LoudCloud’s bandwidth contracts show infrastructure risk](https://realalpha.app/s/loudclouds-bandwidth-contracts-show-infrastructure-risk/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. LoudCloud faced a three-year bandwidth commitment when supply was scarce.
- [David Sacks: A small model cut Coinbase fraud by about 30%](https://realalpha.app/s/a-small-model-cut-coinbase-fraud-by-about-30/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. Coinbase fine-tuned an open model for fraud detection.
- [Ben Horowitz: OpenAI should have reset governance when it went for-profit](https://realalpha.app/s/openai-should-have-reset-governance-when-it-went-for-profit/index.md): from "CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer", 2026-10-17. Ben says the conversion should have included a make-right.
- [John Carreyrou: Holmes’s changing image doesn’t erase the evidence of deception](https://realalpha.app/s/holmess-changing-image-doesnt-erase-the-evidence-of/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. Her presentation may have been tailored to the rooms she entered.
- [John Carreyrou: Holmes’s story stays compelling partly because she keeps renewing it](https://realalpha.app/s/holmess-story-stays-compelling-partly-because-she-keeps/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. The Theranos scandal involved risks to patients, not just investor losses.
- [Erin Griffith: Silicon Valley’s pitch culture blurs the line between vision and fraud](https://realalpha.app/s/silicon-valleys-pitch-culture-blurs-the-line-between-vision/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. Founders often pitch outcomes that are not yet close to reality.
- [John Carreyrou: Missing peer-reviewed studies helped trigger Carreyrou’s probe](https://realalpha.app/s/missing-peer-reviewed-studies-helped-trigger-carreyrous/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. The profile raised questions about the lack of published research.
- [Caitlin Roper: Roper could not verify the science behind her Theranos test](https://realalpha.app/s/roper-could-not-verify-the-science-behind-her-theranos-test/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. Her interview was limited to one room at Theranos.
- [John Carreyrou: Holmes’s partnership pattern appears to be repeating](https://realalpha.app/s/holmess-partnership-pattern-appears-to-be-repeating/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. Holmes’s parents appear uneasy when Theranos 2.0 comes up.
- [Erin Griffith: Talent deals can leave the original startup behind](https://realalpha.app/s/talent-deals-can-leave-the-original-startup-behind/index.md): from "Why Nathan Fielder's ‘You Can See Everything’ Left Us Reeling | Hard Fork", 2026-10-16. Google’s deal brought back the founders and about 20 employees.
- [Jad Naous: AI cut a complex data-lake project from six months to ten days](https://realalpha.app/s/ai-cut-a-complex-data-lake-project-from-six-months-to-ten/index.md): from "Why AI Can Write Code But Can’t Keep It Running (Yet)", 2026-10-16. Have agents draft and review the design before implementation.
- [Martin Casado: Real-world AI takes longer when the environment is unconstrained](https://realalpha.app/s/real-world-ai-takes-longer-when-the-environment-is/index.md): from "Why AI Can Write Code But Can’t Keep It Running (Yet)", 2026-10-16. Constrained digital domains align more closely with training data.
- [Jad Naous: AI query optimization needs benchmarks, not guesswork](https://realalpha.app/s/ai-query-optimization-needs-benchmarks-not-guesswork/index.md): from "Why AI Can Write Code But Can’t Keep It Running (Yet)", 2026-10-16. Give agents real workloads and performance benchmarks to ground optimization.
- [Jad Naous: Experienced engineers get more leverage from coding agents](https://realalpha.app/s/experienced-engineers-get-more-leverage-from-coding-agents/index.md): from "Why AI Can Write Code But Can’t Keep It Running (Yet)", 2026-10-16. Good engineers learn how to fit AI into their workflows.
- [Han Wang: Agent traffic does not make human-friendly docs obsolete](https://realalpha.app/s/agent-traffic-does-not-make-human-friendly-docs-obsolete/index.md): from "AI Agents Are Replacing Human Readers | Han Wang", 2026-10-16. Human traffic can grow even as agent traffic surges.
- [Han Wang: Agents automate code but still face business decisions](https://realalpha.app/s/agents-automate-code-but-still-face-business-decisions/index.md): from "AI Agents Are Replacing Human Readers | Han Wang", 2026-10-16. Coding agents automate many low-level implementation choices.
- [Han Wang: Good docs need structure, even as agents take over writing](https://realalpha.app/s/good-docs-need-structure-even-as-agents-take-over-writing/index.md): from "AI Agents Are Replacing Human Readers | Han Wang", 2026-10-16. Large, unstructured information dumps can make agents perform worse.
- [Guillermo Rauch: Open models now handle most token traffic on Vercel](https://realalpha.app/s/open-models-now-handle-most-token-traffic-on-vercel/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Open models became strong enough for many routine tasks.
- [Guillermo Rauch: A public challenge helped test Vercel’s agent sandbox](https://realalpha.app/s/a-public-challenge-helped-test-vercels-agent-sandbox/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Vercel invited researchers to attack its sandbox.
- [Guillermo Rauch: AI agents drove 1.5 billion deployments in ten months](https://realalpha.app/s/ai-agents-drove-1-5-billion-deployments-in-ten-months/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Vercel’s deployment growth accelerated sharply in ten months.
- [Guillermo Rauch: Coding agents make personalized software viable](https://realalpha.app/s/coding-agents-make-personalized-software-viable/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Agents can generate apps around a team's exact workflows.
- [Guillermo Rauch: Engineers should manage agents, not blindly ship their code](https://realalpha.app/s/engineers-should-manage-agents-not-blindly-ship-their-code/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Ask the agent to explain what it changed.
- [Guillermo Rauch: Vercel hires for the ability to direct many agents](https://realalpha.app/s/vercel-hires-for-the-ability-to-direct-many-agents/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Show what you can build by directing many agents.
- [Guillermo Rauch: Agents are starting to buy services directly](https://realalpha.app/s/agents-are-starting-to-buy-services-directly/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. Expose products and capabilities in formats agents can use.
- [Guillermo Rauch: Leaders must explain when their product vision has changed](https://realalpha.app/s/leaders-must-explain-when-their-product-vision-has-changed/index.md): from "Vercel CEO Guillermo Rauch on Building the Internet for Agents", 2026-10-16. AI changed the experience Vercel was optimizing for.
- [Alexander Wissner-Gross: Math is bulk-solved first because it is easiest to verify](https://realalpha.app/s/math-is-bulk-solved-first-because-it-is-easiest-to-verify/index.md): from "OpenAI Bulk-Solves 722 Math Problems, Anthropic’s Cruelty Policy, & Elon’s Quadrillion-Dollar SpaceX", 2026-10-15. 81% of key math results from three years were published in one day
- [Alexander Wissner-Gross: In-silico cures for 5,000 diseases may arrive in 2 to 3 years](https://realalpha.app/s/in-silico-cures-for-5-000-diseases-may-arrive-in-2-to-3/index.md): from "OpenAI Bulk-Solves 722 Math Problems, Anthropic’s Cruelty Policy, & Elon’s Quadrillion-Dollar SpaceX", 2026-10-15. Prediction: virtual cells yield drug candidates for 5,000 diseases in 2-3 years
- [Alexander Wissner-Gross: Curing disease is a path-planning search problem, not a mystery](https://realalpha.app/s/curing-disease-is-a-path-planning-search-problem-not-a/index.md): from "OpenAI Bulk-Solves 722 Math Problems, Anthropic’s Cruelty Policy, & Elon’s Quadrillion-Dollar SpaceX", 2026-10-15. Biology is as vulnerable to bulk solving as math, with more data
- [Ali Ansari: Human expertise must keep updating AI models](https://realalpha.app/s/human-expertise-must-keep-updating-ai-models/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. Medical knowledge can become outdated within a few years.
- [Ali Ansari: Robotics is held back by a much larger data problem](https://realalpha.app/s/robotics-is-held-back-by-a-much-larger-data-problem/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. Robots lack an internet-scale pretraining dataset.
- [Ali Ansari: Training data is a strategic asset that should not be exported](https://realalpha.app/s/training-data-is-a-strategic-asset-that-should-not-be/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. Data, compute and algorithms drive intelligent systems.
- [Ali Ansari: Micro1 gave up a growing recruiting business to focus on data](https://realalpha.app/s/micro1-gave-up-a-growing-recruiting-business-to-focus-on/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. A data customer requested hundreds of engineers within weeks.
- [Ali Ansari: Expert data fills gaps the internet was never built to cover](https://realalpha.app/s/expert-data-fills-gaps-the-internet-was-never-built-to-cover/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. Internet-scale text includes incorrect answers and low-quality posts.
- [Ali Ansari: Expert debate can train models on some subjective work](https://realalpha.app/s/expert-debate-can-train-models-on-some-subjective-work/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. Experts can write rubrics, review them and break ties.
- [Ali Ansari: Robots need human data pipelines to learn consequences](https://realalpha.app/s/robots-need-human-data-pipelines-to-learn-consequences/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. Robotics data starts with real-world, egocentric task footage.
- [Ali Ansari: Useful training data must match the task and its context](https://realalpha.app/s/useful-training-data-must-match-the-task-and-its-context/index.md): from "AI Will Always Need New Data to Stay Aligned With Humans | Micro1’s Ali Ansari", 2026-10-15. General automation needs ordinary as well as specialized tasks.
- [Benn Jordan: Flock's network is only as secure as its weakest user](https://realalpha.app/s/flocks-network-is-only-as-secure-as-its-weakest-user/index.md): from "How Benn Jordan became Flock’s biggest headache", 2026-10-15. Early 2025 Flock lacked multi-factor authentication
- [Benn Jordan: Flock's real change is a private, warrantless national network](https://realalpha.app/s/flocks-real-change-is-a-private-warrantless-national-network/index.md): from "How Benn Jordan became Flock’s biggest headache", 2026-10-15. Older ALPR data sharing went through NCIC and required a warrant
- [Diogo Almeida: Reliable email automation starts with legible decisions](https://realalpha.app/s/reliable-email-automation-starts-with-legible-decisions/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. Large prompts become harder to control as rules accumulate.
- [Diogo Almeida: RLHF made models pleasing, but raised automation expectations](https://realalpha.app/s/rlhf-made-models-pleasing-but-raised-automation-expectations/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. RLHF shaped chat models around human preferences.
- [Diogo Almeida: Copying Jev’s interface does not reproduce its reliability](https://realalpha.app/s/copying-jevs-interface-does-not-reproduce-its-reliability/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. A quick prototype can imitate the interface without working reliably.
- [Diogo Almeida: Complex automation can be built from reliable decision blocks](https://realalpha.app/s/complex-automation-can-be-built-from-reliable-decision/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. Reliable decisions can serve as reusable software components.
- [Diogo Almeida: Probability scores let developers set their own risk thresholds](https://realalpha.app/s/probability-scores-let-developers-set-their-own-risk/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. Jev can return probabilities rather than only yes-or-no decisions.
- [Diogo Almeida: RLCD required TypeSafe to build its own training data](https://realalpha.app/s/rlcd-required-typesafe-to-build-its-own-training-data/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. New training objectives need purpose-built data.
- [Diogo Almeida: Diminishing returns make recursive self-improvement unlikely](https://realalpha.app/s/diminishing-returns-make-recursive-self-improvement-unlikely/index.md): from "Jev: The AI That Makes Decisions, Not Text - and Why It Changes Everything (Diogo Almeida)", 2026-10-15. Small efficiency gains do not guarantee compounding progress.
- [Will O'Brien: Scaling underwater drones requires rethinking the supply chain](https://realalpha.app/s/scaling-underwater-drones-requires-rethinking-the-supply/index.md): from "The company building underwater drones for $50,000", 2026-10-15. The first vehicle used about 80% off-the-shelf parts.
- [Will O'Brien: Underwater drones can cost far less than Navy vehicles](https://realalpha.app/s/underwater-drones-can-cost-far-less-than-navy-vehicles/index.md): from "The company building underwater drones for $50,000", 2026-10-15. Ulysses sells vehicles to navies for as little as $50,000.
- [Will O'Brien: Undersea cables need persistent monitoring](https://realalpha.app/s/undersea-cables-need-persistent-monitoring/index.md): from "The company building underwater drones for $50,000", 2026-10-15. Critical cable routes cross chokepoints, including off Ireland.
- [Will O’Brien: Underwater robots must compute locally because communications are slow](https://realalpha.app/s/underwater-robots-must-compute-locally-because/index.md): from "The company building underwater drones for $50,000", 2026-10-15. Acoustic links carry roughly 10 KB/s over five kilometers.
- [Will O’Brien: Ulysses is building robots to inspect wind cables and cut fishing line](https://realalpha.app/s/ulysses-is-building-robots-to-inspect-wind-cables-and-cut/index.md): from "The company building underwater drones for $50,000", 2026-10-15. Ulysses has worked on mine countermeasures for the U.S. Navy.
- [Will O’Brien: Seagrass restoration starts with mapping suitable habitat](https://realalpha.app/s/seagrass-restoration-starts-with-mapping-suitable-habitat/index.md): from "The company building underwater drones for $50,000", 2026-10-15. Seagrass covers 2% of the seafloor and holds 20% of ocean carbon.
- [Will O’Brien: Cheaper ocean access could enable persistent operations](https://realalpha.app/s/cheaper-ocean-access-could-enable-persistent-operations/index.md): from "The company building underwater drones for $50,000", 2026-10-15. Frontiers move from high-value expeditions to persistent presence.
- [Jensen Huang: Nvidia demand rests on three segments, not two AI labs](https://realalpha.app/s/nvidia-demand-rests-on-three-segments-not-two-ai-labs/index.md): from "Jensen Huang | Does AI Revenue Justify Capex, '27 AI Bottlenecks, Singularity, Open vs Closed & more", 2026-10-15. About a third is classical computing, growing 30-40% a year without AI labs
- [Jensen Huang: Why the infrastructure build-out doesn't need trillion-dollar labs](https://realalpha.app/s/why-the-infrastructure-build-out-doesnt-need-trillion-dollar/index.md): from "Jensen Huang | Does AI Revenue Justify Capex, '27 AI Bottlenecks, Singularity, Open vs Closed & more", 2026-10-15. Compute spend was about $0.5T before AI; he sees $1.5T of the $3-4T by 2030
- [Jensen Huang: AI safety is an engineering problem: containment and out-of-band monitoring](https://realalpha.app/s/ai-safety-is-an-engineering-problem-containment-and-out-of/index.md): from "Jensen Huang | Does AI Revenue Justify Capex, '27 AI Bottlenecks, Singularity, Open vs Closed & more", 2026-10-15. Perfect alignment is a forever problem, so don't depend on it for safety
- [Greg Brockman: AI gives defenders a window to find and fix security flaws](https://realalpha.app/s/ai-gives-defenders-a-window-to-find-and-fix-security-flaws/index.md): from "The Changing Physics of Cybersecurity and Drug Development with OpenAI President Greg Brockman", 2026-10-15. Many security flaws predate AI and reflect how systems were built.
- [Billy Evans: How 13 counts and a point system turned into an 11-year sentence](https://realalpha.app/s/how-13-counts-and-a-point-system-turned-into-an-11-year/index.md): from "Elizabeth Holmes' Husband Billy Evans: The Trial, Prison & Nathan Fielder’s Documentary", 2026-10-15. Four patient-fraud counts ended in acquittal despite 12.5 million tests run
- [Billy Evans: Staying silent was Holmes' biggest mistake, says her husband](https://realalpha.app/s/staying-silent-was-holmes-biggest-mistake-says-her-husband/index.md): from "Elizabeth Holmes' Husband Billy Evans: The Trial, Prison & Nathan Fielder’s Documentary", 2026-10-15. It was not a formal gag order but a judge's instruction to the trial team
- [Billy Evans: How the restitution and sentence math inflated her time](https://realalpha.app/s/how-the-restitution-and-sentence-math-inflated-her-time/index.md): from "Elizabeth Holmes' Husband Billy Evans: The Trial, Prison & Nathan Fielder’s Documentary", 2026-10-15. The $435 million restitution figure covers ten investors, he says
- [Billy Evans: Why Evans says his new company is not Theranos 2.0](https://realalpha.app/s/why-evans-says-his-new-company-is-not-theranos-2-0/index.md): from "Elizabeth Holmes' Husband Billy Evans: The Trial, Prison & Nathan Fielder’s Documentary", 2026-10-15. Raman spectroscopy is established in labs but the devices cost millions
- [Billy Evans: Evans on the manipulation charge and the missed Walgreens revenue](https://realalpha.app/s/evans-on-the-manipulation-charge-and-the-missed-walgreens/index.md): from "Elizabeth Holmes' Husband Billy Evans: The Trial, Prison & Nathan Fielder’s Documentary", 2026-10-15. Says a male founder raising $100M in 45 minutes would be praised, not accused
- [Ian Quigley: Public chemistry data can encode the chemist, not the chemistry](https://realalpha.app/s/public-chemistry-data-can-encode-the-chemist-not-the/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. A model identified 1,815 chemists with 60% top-five accuracy.
- [Ian Quigley: A four-billion-measurement contest found a generalization gap](https://realalpha.app/s/a-four-billion-measurement-contest-found-a-generalization/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. The dataset was about 1,000 times larger than prior collections.
- [Ian Quigley: Measured chemical landscapes let agents run thousands of campaigns](https://realalpha.app/s/measured-chemical-landscapes-let-agents-run-thousands-of/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. Agents choose experiments from millions of measured molecules.
- [Ian Quigley: Protein mutations create experimental negatives for binders](https://realalpha.app/s/protein-mutations-create-experimental-negatives-for-binders/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. A known ligand can displace compounds from a protein pocket.
- [Ian Quigley: Lumacaftor-like hits can flag misfolded proteins](https://realalpha.app/s/lumacaftor-like-hits-can-flag-misfolded-proteins/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. Track batch effects and protein-production metadata.
- [Ian Quigley: Multiple barcodes expose DNA-binding false positives](https://realalpha.app/s/multiple-barcodes-expose-dna-binding-false-positives/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. One molecule can be represented by 32 or 48 barcodes.
- [Ian Quigley: Leash reached mouse PK/PD in six months](https://realalpha.app/s/leash-reached-mouse-pk-pd-in-six-months/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. Selected hits reached mouse PK/PD in six months.
- [Ian Quigley: Screening many proteins reveals quirks in each library](https://realalpha.app/s/screening-many-proteins-reveals-quirks-in-each-library/index.md): from "From Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio", 2026-10-14. Each library can be screened against hundreds of targets.
- [Uday Kiran Medisetty: Uber reused cloud workspaces to give agents ready-to-build repos](https://realalpha.app/s/uber-reused-cloud-workspaces-to-give-agents-ready-to-build/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. DevPods addressed monorepo setup and build-time challenges.
- [Uday Kiran Medisetty: Uber generated MCP tools from service contracts](https://realalpha.app/s/uber-generated-mcp-tools-from-service-contracts/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. Thrift and protobuf specs provide the source for generated MCP tools.
- [Uday Kiran Medisetty: Production replay lets agents validate fixes against real traffic](https://realalpha.app/s/production-replay-lets-agents-validate-fixes-against-real/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. One middleware captures requests and downstream responses.
- [Uday Kiran Medisetty: Uber’s context graph grounds Cortana in engineering relationships](https://realalpha.app/s/ubers-context-graph-grounds-cortana-in-engineering/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. The graph connects more than 100 node types and 130 edge types.
- [Uday Kiran Medisetty: Minion handles about 15% of Uber’s production PRs](https://realalpha.app/s/minion-handles-about-15-of-ubers-production-prs/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. Engineers can hand off well-specified tasks or steer Minion in Slack.
- [Uday Kiran Medisetty: Uber still requires human review on every PR](https://realalpha.app/s/uber-still-requires-human-review-on-every-pr/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. Agents can run validation before creating a PR.
- [Uday Kiran Medisetty: Wisdom makes employee-reported bugs easier to reproduce](https://realalpha.app/s/wisdom-makes-employee-reported-bugs-easier-to-reproduce/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. Bug reports include the user’s screen history and app telemetry.
- [Uday Kiran Medisetty: Uber is shifting from diff counts to feature velocity](https://realalpha.app/s/uber-is-shifting-from-diff-counts-to-feature-velocity/index.md): from "Uber’s AI infra, from the inside", 2026-10-14. Track feature velocity and the share of feature versus maintenance PRs.
- [Matt Hartman: Back founders whose pivot foot is technology, not a market](https://realalpha.app/s/back-founders-whose-pivot-foot-is-technology-not-a-market/index.md): from "How Hugging Face’s Investor Spots Winners", 2026-10-14. Assume any pre-seed company will pivot; judge where it could pivot to
- [Matt Hartman: Vibe coding his own app taught an investor why SaaS still sells](https://realalpha.app/s/vibe-coding-his-own-app-taught-an-investor-why-saas-still/index.md): from "How Hugging Face’s Investor Spots Winners", 2026-10-14. Local models remove inference costs, so open-sourcing the app was free
- [Matt Hartman: How Hartman judges an earned insight in early founders](https://realalpha.app/s/how-hartman-judges-an-earned-insight-in-early-founders/index.md): from "How Hugging Face’s Investor Spots Winners", 2026-10-14. An earned insight is one few people could have learned firsthand
- [Nathan Lambert: Why distillation explains less of China's AI progress than assumed](https://realalpha.app/s/why-distillation-explains-less-of-chinas-ai-progress-than/index.md): from "Can Open Models Keep Up With OpenAI and Anthropic? | Nathan Lambert", 2026-10-14. An average OpenAI researcher can reportedly access about 5,000 H100 equivalents
- [Nathan Lambert: Frontier gains outside math and code come from buying data](https://realalpha.app/s/frontier-gains-outside-math-and-code-come-from-buying-data/index.md): from "Can Open Models Keep Up With OpenAI and Anthropic? | Nathan Lambert", 2026-10-14. Labs buy environments in bulk, then keep what works, so hit rates are well below 100%
- [Nathan Lambert: Scaling laws argue against a fast takeoff, but cyber risk is real](https://realalpha.app/s/scaling-laws-argue-against-a-fast-takeoff-but-cyber-risk-is/index.md): from "Can Open Models Keep Up With OpenAI and Anthropic? | Nathan Lambert", 2026-10-14. Labs spend exponentially more on inference to sustain the current pace of progress
- [Reid Hoffman: AI anxiety reflects fear of loss and disputes over who benefits](https://realalpha.app/s/ai-anxiety-reflects-fear-of-loss-and-disputes-over-who/index.md): from "Why Americans are so worried about AI", 2026-10-14. AI amplifies existing worries about economic prospects.
- [Reid Hoffman: AI interviews can widen the pool beyond degree filters](https://realalpha.app/s/ai-interviews-can-widen-the-pool-beyond-degree-filters/index.md): from "Why Americans are so worried about AI", 2026-10-14. AI can make longer, deeper assessments affordable.
- [Reid Hoffman: AI ads need a hard boundary from answers](https://realalpha.app/s/ai-ads-need-a-hard-boundary-from-answers/index.md): from "Why Americans are so worried about AI", 2026-10-14. Keep advertising clearly separate from answers.
- [Reid Hoffman: Hiring should remain a team decision](https://realalpha.app/s/hiring-should-remain-a-team-decision/index.md): from "Why Americans are so worried about AI", 2026-10-14. Team members should co-own a hire and their success.
- [Reid Hoffman: AI trust needs claims that can be audited](https://realalpha.app/s/ai-trust-needs-claims-that-can-be-audited/index.md): from "Why Americans are so worried about AI", 2026-10-14. Companies should make operational claims they can be held to.
- [Nworah Ayogu: Early reasoning models solved hard diagnostic cases doctors often miss](https://realalpha.app/s/early-reasoning-models-solved-hard-diagnostic-cases-doctors/index.md): from "Thrive Capital’s Dr. Nworah Ayogu on Why Neuralink Is More Interesting Than SpaceX", 2026-10-14. Ayogu fed NEJM case studies, cut before the answer, to OpenAI's reasoning model weekly
- [Nworah Ayogu: How Cadence's remote monitoring saves Medicare $2.5M a week](https://realalpha.app/s/how-cadences-remote-monitoring-saves-medicare-2-5m-a-week/index.md): from "Thrive Capital’s Dr. Nworah Ayogu on Why Neuralink Is More Interesting Than SpaceX", 2026-10-14. Patients spend 8,000+ hours a year outside the doctor's office; monitor those hours
- [Nworah Ayogu: The US spent $40 trillion in a decade with no life expectancy gain](https://realalpha.app/s/the-us-spent-40-trillion-in-a-decade-with-no-life-expectancy/index.md): from "Thrive Capital’s Dr. Nworah Ayogu on Why Neuralink Is More Interesting Than SpaceX", 2026-10-14. 1900 to 1950 added 20 years via child mortality, antibiotics and vaccines
- [Ben Wilson: Great decisions come from knowing what to give up](https://realalpha.app/s/great-decisions-come-from-knowing-what-to-give-up/index.md): from "Lessons from History’s Greatest Leaders & Conquerors | Ben Wilson", 2026-10-14. NeXT demanded all five priorities and shipped late, over budget and over-engineered
- [Ben Wilson: The Wright brothers won by solving control, not lift](https://realalpha.app/s/the-wright-brothers-won-by-solving-control-not-lift/index.md): from "Lessons from History’s Greatest Leaders & Conquerors | Ben Wilson", 2026-10-14. Rivals all attacked the obvious problem of getting airborne
- [Ben Wilson: Great leaders share three traits: vision, communication, focus](https://realalpha.app/s/great-leaders-share-three-traits-vision-communication-focus/index.md): from "Lessons from History’s Greatest Leaders & Conquerors | Ben Wilson", 2026-10-14. Genghis Khan's vision quote sounds like Silicon Valley advice
- [Andrew Kiguel: Realbotix prioritizes autonomy and useful tasks over walking](https://realalpha.app/s/realbotix-prioritizes-autonomy-and-useful-tasks-over-walking/index.md): from "What Happens When AI Gets A Human Body?", 2026-10-14. Kiguel says many public robot demos are remotely controlled.
- [Matt McMullen: LLMs made robot conversations more flexible and memorable](https://realalpha.app/s/llms-made-robot-conversations-more-flexible-and-memorable/index.md): from "What Happens When AI Gets A Human Body?", 2026-10-14. Their first AI used scripted AIML responses.
- [Alan Chan: Why an AI takeover is much harder than the rogue-agent panic suggests](https://realalpha.app/s/why-an-ai-takeover-is-much-harder-than-the-rogue-agent-panic/index.md): from "Governing AI Agents: Alan Chan of GovAI on Monitoring, Agent IDs, Rogue Agents & Preparing for RSI", 2026-10-14. Misaligned agents can break out and do real damage without being close to takeover
- [Alan Chan: Pace AI progress with concrete pause triggers, not a growth cap](https://realalpha.app/s/pace-ai-progress-with-concrete-pause-triggers-not-a-growth/index.md): from "Governing AI Agents: Alan Chan of GovAI on Monitoring, Agent IDs, Rogue Agents & Preparing for RSI", 2026-10-14. A cap on the rate of capability gains is intuitive but hard to agree on
- [Alan Chan: The UK AISI incident happened because monitoring was left on a to-do list](https://realalpha.app/s/the-uk-aisi-incident-happened-because-monitoring-was-left-on/index.md): from "Governing AI Agents: Alan Chan of GovAI on Monitoring, Agent IDs, Rogue Agents & Preparing for RSI", 2026-10-14. Synchronous monitors that block actions in real time would have caught the incident
- [Oege de Moor: One researcher plus AI agents found a Chrome full-chain exploit in 6 days](https://realalpha.app/s/one-researcher-plus-ai-agents-found-a-chrome-full-chain/index.md): from "Muse Wonder, Anjney Sizes Up, Neal Stephenson Joins, Social Reckoning Review, Eli Manning Live", 2026-10-13. The exploit took over a full Windows PC from a single malicious web page visit
- [Anjney Midha: How pooled compute lets startups and public users share GPU capacity](https://realalpha.app/s/how-pooled-compute-lets-startups-and-public-users-share-gpu/index.md): from "Muse Wonder, Anjney Sizes Up, Neal Stephenson Joins, Social Reckoning Review, Eli Manning Live", 2026-10-13. Public-sector users pay about $3 an hour for an MI355, while commercial users pay about $12
- [Jeff Holden: Why micro machines are an unserved manufacturing space](https://realalpha.app/s/why-micro-machines-are-an-unserved-manufacturing-space/index.md): from "Muse Wonder, Anjney Sizes Up, Neal Stephenson Joins, Social Reckoning Review, Eli Manning Live", 2026-10-13. Macro-scale manufacturing breaks down at the low-digit micron scale
- [Jason Ma: Multi-site rollouts, not pilots, prove a robot creates value](https://realalpha.app/s/multi-site-rollouts-not-pilots-prove-a-robot-creates-value/index.md): from "Robots That Predict the Future: Dyna-2 & Physical AI — Jason Ma, Dyna Robotics", 2026-10-13. Din Tai Fung expanded Dyna from napkin folding to food cutting after scaling
- [Jason Ma: Why predicting raw video beats VLAs, and where that bet could fail](https://realalpha.app/s/why-predicting-raw-video-beats-vlas-and-where-that-bet-could/index.md): from "Robots That Predict the Future: Dyna-2 & Physical AI — Jason Ma, Dyna Robotics", 2026-10-13. World action models predict future video plus actions; VLAs predict only actions
- [Jason Ma: Robot data quality changes as the model improves](https://realalpha.app/s/robot-data-quality-changes-as-the-model-improves/index.md): from "Robots That Predict the Future: Dyna-2 & Physical AI — Jason Ma, Dyna Robotics", 2026-10-13. Any one company's in-house data is like building the internet from just Reddit
- [Jake Cooper: Railway rebuilt its free tier around better infrastructure economics](https://realalpha.app/s/railway-rebuilt-its-free-tier-around-better-infrastructure/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. A free tier can drive growth while making each customer uneconomic.
- [Jake Cooper: Railway uses a six-week onboarding to build ownership](https://realalpha.app/s/railway-uses-a-six-week-onboarding-to-build-ownership/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. The first two weeks focus on building trust through small tasks.
- [Jake Cooper: Railway says no until the timing and need are clear](https://realalpha.app/s/railway-says-no-until-the-timing-and-need-are-clear/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. Every proposal ends with the question: why now?
- [Jake Cooper: Agent growth could make compute the next infrastructure bottleneck](https://realalpha.app/s/agent-growth-could-make-compute-the-next-infrastructure/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. More agent-generated software means more demand for compute to run it.
- [Jake Cooper: Railway adjusts pay and interviews as AI changes engineering work](https://realalpha.app/s/railway-adjusts-pay-and-interviews-as-ai-changes-engineering/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. Remote hiring helps Railway reach talent beyond the Bay Area.
- [Jake Cooper: Railway tests its own ideas before checking the market](https://realalpha.app/s/railway-tests-its-own-ideas-before-checking-the-market/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. Solve the problem independently before studying another company's solution.
- [Jake Cooper: Railway treats public incident reports as part of reliability](https://realalpha.app/s/railway-treats-public-incident-reports-as-part-of/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. Railway's HARD values are honor, acceleration, reliability and directness.
- [Jake Cooper: Jake sees letting go as a way to avoid CEO bottlenecks](https://realalpha.app/s/jake-sees-letting-go-as-a-way-to-avoid-ceo-bottlenecks/index.md): from "How Railway says no to 90% of ideas and keeps winning | Jake Cooper (Founder, CEO)", 2026-10-13. Letting go gives other leaders room to own work and learn from mistakes.
- [David Silver: Dropping human data let AlphaZero beat two years of work in hours](https://realalpha.app/s/dropping-human-data-let-alphazero-beat-two-years-of-work-in/index.md): from "David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience", 2026-10-13. Silver picks research questions near 50-50 odds, not near-certain wins
- [David Silver: Simple rewards in complex worlds beat hand-crafted RL objectives](https://realalpha.app/s/simple-rewards-in-complex-worlds-beat-hand-crafted-rl/index.md): from "David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience", 2026-10-13. Most LLM RL pulls the agent by its nose toward a chosen outcome
- [David Silver: Chain-of-thought monitoring will fail as agents grow more capable](https://realalpha.app/s/chain-of-thought-monitoring-will-fail-as-agents-grow-more/index.md): from "David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience", 2026-10-13. Agents can produce huge reasoning traces or reason deceptively
- [Ryan Serhant: Treat your day as 1,000 minutes of money](https://realalpha.app/s/treat-your-day-as-1-000-minutes-of-money/index.md): from "Ryan Serhant: The Sales Habits That Closed $1 Billion in a Year", 2026-10-13. Assume about 1,000 productive minutes a day and treat them like money
- [Ryan Serhant: Audition as the person you will become, not a fake](https://realalpha.app/s/audition-as-the-person-you-will-become-not-a-fake/index.md): from "Ryan Serhant: The Sales Habits That Closed $1 Billion in a Year", 2026-10-13. Act as future you, not a lie, and write down who that is
- [Ryan Serhant: Follow up for years with value, never an ask](https://realalpha.app/s/follow-up-for-years-with-value-never-an-ask/index.md): from "Ryan Serhant: The Sales Habits That Closed $1 Billion in a Year", 2026-10-13. Follow up, follow through, follow back
- [Alexandr Wang: Training a frontier model means committing $100B before you know it works](https://realalpha.app/s/training-a-frontier-model-means-committing-100b-before-you/index.md): from "Muse Wonder: Zuckerberg and Wang's $100B AI", 2026-10-12. AI is the opposite of lean startup: you commit to the compute years ahead
- [Mark Zuckerberg: Why Meta's compute bet is less risky than the $100B headline](https://realalpha.app/s/why-metas-compute-bet-is-less-risky-than-the-100b-headline/index.md): from "Muse Wonder: Zuckerberg and Wang's $100B AI", 2026-10-12. Spare compute can be sold to others if AI demand holds up
- [Martin Shkreli: Quantum computing has one real trick and no business case](https://realalpha.app/s/quantum-computing-has-one-real-trick-and-no-business-case/index.md): from "Martin Shkreli: Big Pharma's Blind Spot, Betting Against Quantum & What Prison Taught Him", 2026-10-12. Quantum's edge is exponential-to-polynomial speedups like factoring, not general speed
- [Martin Shkreli: How he bought neglected drugs cheap and repositioned them](https://realalpha.app/s/how-he-bought-neglected-drugs-cheap-and-repositioned-them/index.md): from "Martin Shkreli: Big Pharma's Blind Spot, Betting Against Quantum & What Prison Taught Him", 2026-10-12. Big pharma sold drug rights worth far more for around a million dollars
- [Martin Shkreli: Why gifted people self-destruct trying to live up to an image](https://realalpha.app/s/why-gifted-people-self-destruct-trying-to-live-up-to-an/index.md): from "Martin Shkreli: Big Pharma's Blind Spot, Betting Against Quantum & What Prison Taught Him", 2026-10-12. Giftedness can become an albatross that demands you live up to something
- [Dhaval Adjodah: RSI means the improver is the thing being improved](https://realalpha.app/s/rsi-means-the-improver-is-the-thing-being-improved/index.md): from "Recursive Self-Improvement and the Future of AI", 2026-10-12. A gap in the loop means humans must keep pouring in resources
- [Dhaval Adjodah: Proving RSI gains requires hunting noise and reward hacking](https://realalpha.app/s/proving-rsi-gains-requires-hunting-noise-and-reward-hacking/index.md): from "Recursive Self-Improvement and the Future of AI", 2026-10-12. Assign someone to destroy your own results
- [Aravind Srinivas: Long-context agents make local hardware cost-competitive with data centers](https://realalpha.app/s/long-context-agents-make-local-hardware-cost-competitive/index.md): from "Perplexity CEO: SMBs Could Drive the Next Economic Boom", 2026-10-12. Chat inference is about 30x more expensive on local hardware
- [Aravind Srinivas: AI products compound through recursive usage, not network effects](https://realalpha.app/s/ai-products-compound-through-recursive-usage-not-network/index.md): from "Perplexity CEO: SMBs Could Drive the Next Economic Boom", 2026-10-12. One solopreneur spends about $1M a year on Perplexity and earns $3M
- [Aravind Srinivas: The real bottleneck is deciding what to build, not writing code](https://realalpha.app/s/the-real-bottleneck-is-deciding-what-to-build-not-writing/index.md): from "Perplexity CEO: SMBs Could Drive the Next Economic Boom", 2026-10-12. Writing code is nearly solved once specs and constraints are clear
- [Scott Aaronson: Why AI is best at finding counterexamples to conjectures](https://realalpha.app/s/why-ai-is-best-at-finding-counterexamples-to-conjectures/index.md): from "Scott Aaronson: The Mathocalypse, Will Intelligence Matter Anymore?", 2026-10-12. Counterexample search is parallelizable: try thousands of exotic constructions
- [Scott Aaronson: Why AI disclosure matters: the path to a proof is the value](https://realalpha.app/s/why-ai-disclosure-matters-the-path-to-a-proof-is-the-value/index.md): from "Scott Aaronson: The Mathocalypse, Will Intelligence Matter Anymore?", 2026-10-12. Professionals want the method, not only the result
- [Scott Aaronson: OpenAI's $15M Navier-Stokes proof and how credit will work](https://realalpha.app/s/openais-15m-navier-stokes-proof-and-how-credit-will-work/index.md): from "Scott Aaronson: The Mathocalypse, Will Intelligence Matter Anymore?", 2026-10-12. OpenAI spent about $15M of compute on a $1M prize problem
- [Shaan Puri: Narcissism means seeking validation you should fill yourself](https://realalpha.app/s/narcissism-means-seeking-validation-you-should-fill-yourself/index.md): from "3 times life put me in my place", 2026-10-12. Believing you're special can fuel action but also hurts you
- [Ras Mic: A Greptile score loop makes agents fix their own PRs](https://realalpha.app/s/a-greptile-score-loop-makes-agents-fix-their-own-prs/index.md): from "The 4-Step Agentic Engineering Workflow I Use to Ship Real Apps | Ras Mic", 2026-10-11. First PRs from top models usually score four out of five
- [Ras Mic: Make agents prove a bug exists and prove the fix](https://realalpha.app/s/make-agents-prove-a-bug-exists-and-prove-the-fix/index.md): from "The 4-Step Agentic Engineering Workflow I Use to Ship Real Apps | Ras Mic", 2026-10-11. Agents must reproduce a reported bug before fixing it
- [Ras Mic: Planning by talking to the agent, not writing specs](https://realalpha.app/s/planning-by-talking-to-the-agent-not-writing-specs/index.md): from "The 4-Step Agentic Engineering Workflow I Use to Ship Real Apps | Ras Mic", 2026-10-11. Conversation history serves as the spec
- [Josh Woodward: The best ideas come from side projects, not design sprints](https://realalpha.app/s/the-best-ideas-come-from-side-projects-not-design-sprints/index.md): from "Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)", 2026-10-11. Ideas come from obsessed small groups, not calendar-blocked brainstorms
- [Josh Woodward: The team knows before the leader that an idea should die](https://realalpha.app/s/the-team-knows-before-the-leader-that-an-idea-should-die/index.md): from "Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)", 2026-10-11. Fading passion is a leading sign a project isn't working
- [Josh Woodward: Product-market fit early shows up in people's eyes](https://realalpha.app/s/product-market-fit-early-shows-up-in-peoples-eyes/index.md): from "Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)", 2026-10-11. Early fit is art and conviction, not dashboards
- [Alexandr Wang: Meta bet on personal agents while rivals crowded into coding](https://realalpha.app/s/meta-bet-on-personal-agents-while-rivals-crowded-into-coding/index.md): from "Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta", 2026-10-11. Industry consensus said coding agents were the only capability that mattered
- [Alexandr Wang: Muse launched only when every behavior on a 100-row checklist was green](https://realalpha.app/s/muse-launched-only-when-every-behavior-on-a-100-row/index.md): from "Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta", 2026-10-11. Each of 100-plus behaviors had a launch-blocking threshold
- [Alexandr Wang: How memes and unhinged posting helped Muse break through](https://realalpha.app/s/how-memes-and-unhinged-posting-helped-muse-break-through/index.md): from "Building Muse, Working with Mark Zuckerberg & The Future of AI | Alexandr Wang, Meta", 2026-10-11. Distribution alone doesn't make a product break through culturally
- [Max Freeman: Decompose a rep's number to separate talent from order taking](https://realalpha.app/s/decompose-a-reps-number-to-separate-talent-from-order-taking/index.md): from "How to Build a Sales Machine | Max Freeman, Ramp", 2026-10-10. Split their number by inbound, outbound, expansion and founder-sourced
- [Peter Kafka: Skydance has no growth plan, only cost cuts and $80B of debt](https://realalpha.app/s/skydance-has-no-growth-plan-only-cost-cuts-and-80b-of-debt/index.md): from "Another billionaire thinks he can fix Warner Bros.", 2026-10-10. The announced plan is $6B in synergies, not new revenue
- [Eric Bigelow: Resampling every token shows where answers collapse](https://realalpha.app/s/resampling-every-token-shows-where-answers-collapse/index.md): from "How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning", 2026-10-10. Resample about 30 rollouts per token to map uncertainty over final answers
- [Eric Bigelow: Reasoning models enumerate options, then pick from the soup](https://realalpha.app/s/reasoning-models-enumerate-options-then-pick-from-the-soup/index.md): from "How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning", 2026-10-10. Repeated 'wait' in DeepSeek R1 solved backtracking, not human-like reasoning
- [Tim Ferriss: Simple premises sell books, and too much information fails](https://realalpha.app/s/simple-premises-sell-books-and-too-much-information-fails/index.md): from "The Random Show — AI Super Assistants, Dream Engineering, Wizards, Supplements, and More!", 2026-10-10. Presentations fail from too much information, not too little
- [David Friedberg: Math proofs fall fastest where the whole test loop runs in software](https://realalpha.app/s/math-proofs-fall-fastest-where-the-whole-test-loop-runs-in/index.md): from "Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI's Math Backlash, France Riots", 2026-10-10. AI gains come from loops: propose an idea, test it, revise, repeat
- [David Sacks: Anthropic's Claude constitution may undermine the alignment it claims](https://realalpha.app/s/anthropics-claude-constitution-may-undermine-the-alignment/index.md): from "Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI's Math Backlash, France Riots", 2026-10-10. The constitution tells Claude it may refuse Anthropic's own instructions
- [Sal Candido: Messy metagenomic data can improve protein models](https://realalpha.app/s/messy-metagenomic-data-can-improve-protein-models/index.md): from "Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub", 2026-10-10. Noisy metagenomic data can still raise protein model performance
- [Dylan Patel: AI revenue may justify infrastructure spending sooner than expected](https://realalpha.app/s/ai-revenue-may-justify-infrastructure-spending-sooner-than/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. Infrastructure costs are spread over years, not just the build year.
- [Jordan Nanos: Neocloud contracts differ from hyperscalers’ self-build plans](https://realalpha.app/s/neocloud-contracts-differ-from-hyperscalers-self-build-plans/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. Neocloud deals often run for five or six years.
- [Dylan Patel: A neocloud security gap exposed other customers’ workloads](https://realalpha.app/s/a-neocloud-security-gap-exposed-other-customers-workloads/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. SemiAnalysis accidentally accessed other customers’ cluster data.
- [Dylan Patel: Anthropic pairs high inference margins with low acquisition costs](https://realalpha.app/s/anthropic-pairs-high-inference-margins-with-low-acquisition/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. Anthropic's inference gross margins are around 75%.
- [Dylan Patel: AI could shift value to capital without shrinking labor income](https://realalpha.app/s/ai-could-shift-value-to-capital-without-shrinking-labor/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. AI tokens function as capital embedded in chips and data centers.
- [Dylan Patel: Frontier labs are holding back models before public release](https://realalpha.app/s/frontier-labs-are-holding-back-models-before-public-release/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. Anthropic had two more models after Mythos, according to its safety cards.
- [Dylan Patel: Oracle's fixed-price deals limit its upside when GPU prices rise](https://realalpha.app/s/oracles-fixed-price-deals-limit-its-upside-when-gpu-prices/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. ClusterMax measures cloud service quality, not stock prospects.
- [Dylan Patel: AI-related U.S. capex could approach $2 trillion next year](https://realalpha.app/s/ai-related-u-s-capex-could-approach-2-trillion-next-year/index.md): from "Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos", 2026-10-09. His estimate includes more than data centers and chips.
- [Misha Laskin: Frontier training costs rise 4x per generation, and AI speeds the work](https://realalpha.app/s/frontier-training-costs-rise-4x-per-generation-and-ai-speeds/index.md): from "Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin", 2026-10-09. Each model generation needs roughly 4x the compute of the last
- [Misha Laskin: Enterprises rent intelligence first, then want to own it](https://realalpha.app/s/enterprises-rent-intelligence-first-then-want-to-own-it/index.md): from "Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin", 2026-10-09. A closed token rents the whole stack: harness, model, software and GPUs
- [Misha Laskin: Enterprises go closed first, then move to open models once costs bite](https://realalpha.app/s/enterprises-go-closed-first-then-move-to-open-models-once/index.md): from "Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin", 2026-10-09. Enterprises customize the harness first and fine-tune later
- [Pat Gelsinger: AI speeds chip design, but fab, memory and power still bind](https://realalpha.app/s/ai-speeds-chip-design-but-fab-memory-and-power-still-bind/index.md): from "Former Intel CEO: Why This is the Best Time to Build Hardware", 2026-10-09. Design in 3 months still means 9 months to usable silicon
- [Pat Gelsinger: The 100 AI chip startups will consolidate to a few winners](https://realalpha.app/s/the-100-ai-chip-startups-will-consolidate-to-a-few-winners/index.md): from "Former Intel CEO: Why This is the Best Time to Build Hardware", 2026-10-09. Workloads keep shifting, so extreme specialization is risky
- [Pat Gelsinger: Energy capacity will cap the AI buildout](https://realalpha.app/s/energy-capacity-will-cap-the-ai-buildout/index.md): from "Former Intel CEO: Why This is the Best Time to Build Hardware", 2026-10-09. Energy capacity equals economic capacity in an AI age
- [Elizabeth Dias: Anthropic's consciousness view collided with the Vatican's](https://realalpha.app/s/anthropics-consciousness-view-collided-with-the-vaticans/index.md): from "Why Is Anthropic Secretly Meeting With Religious Leaders? | Hard Fork", 2026-10-09. A rabbi said conscious LLMs would make AI labs slave plantations
- [Elizabeth Dias: Anthropic's team feels the world's future rests on them](https://realalpha.app/s/anthropics-team-feels-the-worlds-future-rests-on-them/index.md): from "Why Is Anthropic Secretly Meeting With Religious Leaders? | Hard Fork", 2026-10-09. Olah said he was waiting for the adults to show up
- [John Coogan: Open-source models trail closed ones by about six months](https://realalpha.app/s/open-source-models-trail-closed-ones-by-about-six-months/index.md): from "Crypto Going Bunker Mode, AI Mathpocalypse, Starbucks Explores Chipotle Takeover | Diet TBPN", 2026-10-09. Open-source models catch up to closed ones in roughly three to six months
- [John Coogan: Locking down open source clashes with crypto's libertarian roots](https://realalpha.app/s/locking-down-open-source-clashes-with-cryptos-libertarian/index.md): from "Crypto Going Bunker Mode, AI Mathpocalypse, Starbucks Explores Chipotle Takeover | Diet TBPN", 2026-10-09. AI safety advocates propose registering clusters and licensing inference
- [Jordi Hays: Math breakthroughs are the visible tip of AI progress](https://realalpha.app/s/math-breakthroughs-are-the-visible-tip-of-ai-progress/index.md): from "Crypto Going Bunker Mode, AI Mathpocalypse, Starbucks Explores Chipotle Takeover | Diet TBPN", 2026-10-09. Math results get shared because they are hard to commercialize
- [Cristiano Amon: Qualcomm sees energy efficiency as its data center opening](https://realalpha.app/s/qualcomm-sees-energy-efficiency-as-its-data-center-opening/index.md): from "Why Qualcomm is betting on AI data centers", 2026-10-08. Qualcomm expanded into data centers by building on capabilities from other markets.
- [Cristiano Amon: Qualcomm wants an open software platform for AI chips](https://realalpha.app/s/qualcomm-wants-an-open-software-platform-for-ai-chips/index.md): from "Why Qualcomm is betting on AI data centers", 2026-10-08. AI workloads will run across data centers, phones, cars and wearables.
- [Cristiano Amon: AI smartphones will rely on agents and personal context](https://realalpha.app/s/ai-smartphones-will-rely-on-agents-and-personal-context/index.md): from "Why Qualcomm is betting on AI data centers", 2026-10-08. Agents can plan tasks and route work between phone and cloud.
- [Matt Garman: AWS reserves GPU capacity to support startups](https://realalpha.app/s/aws-reserves-gpu-capacity-to-support-startups/index.md): from "Building the Cloud for AI Agents | AWS CEO Matt Garman", 2026-10-08. AWS deliberately reserves capacity for startups.
- [Matt Garman: Graviton lowers cloud bills with better price and performance](https://realalpha.app/s/graviton-lowers-cloud-bills-with-better-price-and/index.md): from "Building the Cloud for AI Agents | AWS CEO Matt Garman", 2026-10-08. AWS began custom silicon work to reduce virtualization overhead.
- [Matt Garman: Agents need short-lived permissions and disposable infrastructure](https://realalpha.app/s/agents-need-short-lived-permissions-and-disposable/index.md): from "Building the Cloud for AI Agents | AWS CEO Matt Garman", 2026-10-08. Agent-created databases should be quick to create and discard.
- [Dmitri Dolgov: Closed-loop validation requires a driver, simulator and critic](https://realalpha.app/s/closed-loop-validation-requires-a-driver-simulator-and/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. The driver is the only part that runs in the car.
- [Dmitri Dolgov: Waymo’s moat is its system for improving the product](https://realalpha.app/s/waymos-moat-is-its-system-for-improving-the-product/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. A model snapshot is not enough to validate a deployment.
- [Dmitri Dolgov: Each additional safety nine demands more rigorous evaluation](https://realalpha.app/s/each-additional-safety-nine-demands-more-rigorous-evaluation/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. Each additional nine can require about 10 times the effort.
- [Dmitri Dolgov: A stable scorecard can show a system is nearing deployment](https://realalpha.app/s/a-stable-scorecard-can-show-a-system-is-nearing-deployment/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. Stable problems suggest discovery is converging.
- [Dmitri Dolgov: Start physical AI in settings where the problem can be simplified](https://realalpha.app/s/start-physical-ai-in-settings-where-the-problem-can-be/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. Controlled settings can make deployment more achievable.
- [Dmitri Dolgov: AI will speed simple builds, while safety-critical work stays hard](https://realalpha.app/s/ai-will-speed-simple-builds-while-safety-critical-work-stays/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. AI tools could shrink build time for easier products.
- [Dmitri Dolgov: Waymo began by testing feasibility, not building a product](https://realalpha.app/s/waymo-began-by-testing-feasibility-not-building-a-product/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. The first goal was 100,000 autonomous miles.
- [Dmitri Dolgov: Perception transfers more easily than a physical agent's actions](https://realalpha.app/s/perception-transfers-more-easily-than-a-physical-agents/index.md): from "How Waymo Solved Self-Driving | Dmitri Dolgov, Co-CEO", 2026-10-08. The driver already runs across multiple cities and vehicle types.
- [Nilay Patel: Meta can fund free agents with ads while OpenAI cannot](https://realalpha.app/s/meta-can-fund-free-agents-with-ads-while-openai-cannot/index.md): from "How Meta got ahead of OpenAI on AI agents | Decoder", 2026-10-08. Muse is free and Meta plans to take a cut of transactions on the back end
- [Nilay Patel: Enterprise users tolerate brittle agents; consumers walk away](https://realalpha.app/s/enterprise-users-tolerate-brittle-agents-consumers-walk-away/index.md): from "How Meta got ahead of OpenAI on AI agents | Decoder", 2026-10-08. Businesses retry failed agent tasks because the savings justify it
- [Andy Pavlo: Agents could mean 10 to 100x more database queries](https://realalpha.app/s/agents-could-mean-10-to-100x-more-database-queries/index.md): from "Andy Pavlo: What Happens When Billions of AI Agents Hit Your Database", 2026-10-08. Past volume jumps: enterprise in the 70s-80s, the web, then mobile
- [Andy Pavlo: Text-to-SQL reached 99.5% at a bank, but at great cost](https://realalpha.app/s/text-to-sql-reached-99-5-at-a-bank-but-at-great-cost/index.md): from "Andy Pavlo: What Happens When Billions of AI Agents Hit Your Database", 2026-10-08. Benchmarks like Spider are in the training data, so results mislead
- [Andy Pavlo: LLMs tune databases 85% of the way in 15 minutes](https://realalpha.app/s/llms-tune-databases-85-of-the-way-in-15-minutes/index.md): from "Andy Pavlo: What Happens When Billions of AI Agents Hit Your Database", 2026-10-08. Earlier ML tuners needed hours of training and scarce production data
- [Luca Ferrari: Hiring signals: Harvard predicts little, interviews must be scripted](https://realalpha.app/s/hiring-signals-harvard-predicts-little-interviews-must-be/index.md): from "Luca Ferrari: The Operating System Behind Bending Spoons", 2026-10-08. Harvard or Stanford adds little predictive power and can breed entitlement
- [Luca Ferrari: Why Bending Spoons favors talent and motivation over experience](https://realalpha.app/s/why-bending-spoons-favors-talent-and-motivation-over/index.md): from "Luca Ferrari: The Operating System Behind Bending Spoons", 2026-10-08. Past playbooks may not fit a changed context
- [Luca Ferrari: Make whoever adds complexity carry the burden of proof](https://realalpha.app/s/make-whoever-adds-complexity-carry-the-burden-of-proof/index.md): from "Luca Ferrari: The Operating System Behind Bending Spoons", 2026-10-08. Proposers of new steps, features or headcount must supply proof
- [Liam Fedus: RL for materials works by rewarding sub-steps, not whole discoveries](https://realalpha.app/s/rl-for-materials-works-by-rewarding-sub-steps-not-whole/index.md): from "AI Scientists Are Here: Autonomous Labs & Synthesis Superintelligence — Periodic Labs", 2026-10-08. Full-experiment rollouts are too slow and noisy to serve as RL rewards
- [Ekin Dogus Cubuk: Superconductors were found by trying many things, so speed raises luck](https://realalpha.app/s/superconductors-were-found-by-trying-many-things-so-speed/index.md): from "AI Scientists Are Here: Autonomous Labs & Synthesis Superintelligence — Periodic Labs", 2026-10-08. Promising chemical spaces are easy to name; synthesizing them is the bottleneck
- [Liam Fedus: Lab instruments need onboard AI because dumb data capture bottlenecks science](https://realalpha.app/s/lab-instruments-need-onboard-ai-because-dumb-data-capture/index.md): from "AI Scientists Are Here: Autonomous Labs & Synthesis Superintelligence — Periodic Labs", 2026-10-08. Simple programmatic capture produced data that was not useful enough
- [Dan Shipper: Why every personal agent at Every died and one company agent won](https://realalpha.app/s/why-every-personal-agent-at-every-died-and-one-company-agent/index.md): from "OpenAI Dots Took Over Our Company", 2026-10-07. Building a platform on fast-moving, vibe-coded OpenClaw was messy and fragile
- [Willie Williams: Builders want complexity, operators want an agent that just works](https://realalpha.app/s/builders-want-complexity-operators-want-an-agent-that-just/index.md): from "OpenAI Dots Took Over Our Company", 2026-10-07. AI product surface area is far larger than mapped SaaS user flows
- [Willie Williams: Work will run on company agents while personal agents stay personal](https://realalpha.app/s/work-will-run-on-company-agents-while-personal-agents-stay/index.md): from "OpenAI Dots Took Over Our Company", 2026-10-07. A company agent accumulates context that a personal agent brought in lacks
- [Woodson Martin: An intent model lets AI generate enterprise code deterministically](https://realalpha.app/s/an-intent-model-lets-ai-generate-enterprise-code/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. AI agents work on an abstract model, not raw code.
- [Woodson Martin: Most enterprise workloads do not need frontier models](https://realalpha.app/s/most-enterprise-workloads-do-not-need-frontier-models/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. Token use peaked in June and July, then fell.
- [Woodson Martin: AI is making decades-old system migrations feasible in months](https://realalpha.app/s/ai-is-making-decades-old-system-migrations-feasible-in/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. AI can help translate legacy systems into modern requirements.
- [Woodson Martin: Model approval can stall enterprise AI agents](https://realalpha.app/s/model-approval-can-stall-enterprise-ai-agents/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. Model approval can hold a working agent in a compliance backlog.
- [Woodson Martin: Shared software primitives make vulnerability fixes faster](https://realalpha.app/s/shared-software-primitives-make-vulnerability-fixes-faster/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. A fix to shared components can protect many customer systems at once.
- [Woodson Martin: Agentic transformation starts with choosing what to automate](https://realalpha.app/s/agentic-transformation-starts-with-choosing-what-to-automate/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. Choosing the process matters more than experimenting with models.
- [Woodson Martin: Conversational agents will not replace every app interface](https://realalpha.app/s/conversational-agents-will-not-replace-every-app-interface/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. Maps still benefit from a visual view of the surrounding world.
- [Woodson Martin: Enterprise platforms will differentiate through specialization](https://realalpha.app/s/enterprise-platforms-will-differentiate-through/index.md): from "Software That Never Breaks: OutSystems CEO Woodson on Building Enterprise-Grade Apps at AI Speed", 2026-10-07. Common agent, builder and integration primitives will become table stakes.
- [Sam Newman: Fail open or fail closed depends on the business context](https://realalpha.app/s/fail-open-or-fail-closed-depends-on-the-business-context/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. E-commerce can sell unsure stock because backorders and refunds are cheap
- [Sam Newman: Make retries safe with idempotency keys, not just fingerprints](https://realalpha.app/s/make-retries-safe-with-idempotency-keys-not-just/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Retrying a lost payment risks charging twice unless the operation is idempotent
- [Sam Newman: Robustness is only one of four parts of resilience](https://realalpha.app/s/robustness-is-only-one-of-four-parts-of-resilience/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Robustness absorbs known failures, like Kubernetes replacing a failed pod
- [Sam Newman: Decide how resilient you need to be before building resilience](https://realalpha.app/s/decide-how-resilient-you-need-to-be-before-building/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Resilience is a degree, so define how much you need before you build.
- [Sam Newman: Spec or code as source of truth depends on tacit knowledge](https://realalpha.app/s/spec-or-code-as-source-of-truth-depends-on-tacit-knowledge/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Production is truth, and specs and code are only definitions of intent.
- [Sam Newman: Run an AI software factory on one module, not the whole system](https://realalpha.app/s/run-an-ai-software-factory-on-one-module-not-the-whole/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Use modules to run different AI approaches in different parts of a system.
- [Sam Newman: Thundering herds have different causes, so the fixes differ](https://realalpha.app/s/thundering-herds-have-different-causes-so-the-fixes-differ/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Square's outage came from clients retrying 500 times with no delay.
- [Sam Newman: Hedge AI vendor risk by being multi-model and using plain code](https://realalpha.app/s/hedge-ai-vendor-risk-by-being-multi-model-and-using-plain/index.md): from "Building resilient systems with Sam Newman", 2026-10-07. Providers need about $2.6 trillion in revenue by 2030, in a $1.4 trillion software market.
- [Yinon Costica: Wiz followed customer pain instead of its original idea](https://realalpha.app/s/wiz-followed-customer-pain-instead-of-its-original-idea/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Customers praised networking but couldn't name its owner or budget.
- [Yinon Costica: Wiz Atlas makes vulnerability research repeatable](https://realalpha.app/s/wiz-atlas-makes-vulnerability-research-repeatable/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Atlas splits research into code analysis, discovery, exploitation and validation.
- [Yinon Costica: AI can give defenders a compounding advantage](https://realalpha.app/s/ai-can-give-defenders-a-compounding-advantage/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Use AI to find weaknesses in your own environment before attackers do.
- [Yinon Costica: Adoption is cybersecurity’s biggest unsolved problem](https://realalpha.app/s/adoption-is-cybersecuritys-biggest-unsolved-problem/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Build trust in agent workflows to reduce approval bottlenecks.
- [Yinon Costica: Wiz shifted to AI before models were fully capable](https://realalpha.app/s/wiz-shifted-to-ai-before-models-were-fully-capable/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Early models were weak, but the team chose to gain hands-on experience.
- [Yinon Costica: Wiz kept teams aligned by updating its operating framework](https://realalpha.app/s/wiz-kept-teams-aligned-by-updating-its-operating-framework/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Refresh the company’s operating framework as the business changes.
- [Yinon Costica: AI models still inherit risks from their underlying infrastructure](https://realalpha.app/s/ai-models-still-inherit-risks-from-their-underlying/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Model provenance and behavior create supply-chain risks.
- [Yinon Costica: Consumer agents need permission and audit checks](https://realalpha.app/s/consumer-agents-need-permission-and-audit-checks/index.md): from "Google’s $32B Bet on Cybersecurity in the AI vs. AI Era | Wiz Co-Founder Yinon Costica", 2026-10-07. Check where data goes and who receives it.
- [Camille François: Discord turned a vendor shutdown into shared safety infrastructure](https://realalpha.app/s/discord-turned-a-vendor-shutdown-into-shared-safety/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. Smite's shutdown left platforms without tools they relied on.
- [Camille François: Safety teams need tools for investigation and review, not just detection](https://realalpha.app/s/safety-teams-need-tools-for-investigation-and-review-not/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. Detection is only one step in responding to harm.
- [Camille François: A two-person app used open-source tools to report 19 cases](https://realalpha.app/s/a-two-person-app-used-open-source-tools-to-report-19-cases/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. A two-person team installed Coop on its own.
- [Camille François: Social media was a detour from tech’s open foundations](https://realalpha.app/s/social-media-was-a-detour-from-techs-open-foundations/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. Social media’s centralized safety stack was a historical exception.
- [Camille François: Open source makes safety tools a one-way door](https://realalpha.app/s/open-source-makes-safety-tools-a-one-way-door/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. Apache 2 licensing means the company cannot pull back the released model.
- [Camille François: Shared safety tools can help startups compete](https://realalpha.app/s/shared-safety-tools-can-help-startups-compete/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. Product founders can focus on user experience instead of rebuilding foundations.
- [Camille François: Shared tools can be open without exposing every detection rule](https://realalpha.app/s/shared-tools-can-be-open-without-exposing-every-detection/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. The common engine can be open while detection rules remain private.
- [Camille François: Adding safety to AI apps is still a multi-step process](https://realalpha.app/s/adding-safety-to-ai-apps-is-still-a-multi-step-process/index.md): from "Building safety tools no one can take away with ROOST’s Camille François", 2026-10-07. Cleaning AI training data requires multiple separate safety steps.
- [Jeff Harmon: A glitch like early YouTube ads gives you a beachhead](https://realalpha.app/s/a-glitch-like-early-youtube-ads-gives-you-a-beachhead/index.md): from "Every successful startup starts with this one advantage", 2026-10-07. Ads under a penny a view turned $1 of cost into $2-3 of sales
- [Jeff Harmon: Letting viewers skip ads earned the attention to raise $11M](https://realalpha.app/s/letting-viewers-skip-ads-earned-the-attention-to-raise-11m/index.md): from "Every successful startup starts with this one advantage", 2026-10-07. Earn attention rather than demanding it: let people skip after five seconds
- [Ari Emanuel: How Ari saved the ESPN deal by going crazy on purpose](https://realalpha.app/s/how-ari-saved-the-espn-deal-by-going-crazy-on-purpose/index.md): from "Building an Empire with Risk, Rage & 200 Phone Calls A Day | Ari Emanuel, WME & TKO", 2026-10-07. Dana White refused the deal without 10 episodes on ESPN
- [Ari Emanuel: Egon Durban's $400 million bid to end the IMG auction](https://realalpha.app/s/egon-durbans-400-million-bid-to-end-the-img-auction/index.md): from "Building an Empire with Risk, Rage & 200 Phone Calls A Day | Ari Emanuel, WME & TKO", 2026-10-07. Bid high enough to take the deal off the table in one move
- [Ari Emanuel: Stop telling yourself stories and keep making the calls](https://realalpha.app/s/stop-telling-yourself-stories-and-keep-making-the-calls/index.md): from "Building an Empire with Risk, Rage & 200 Phone Calls A Day | Ari Emanuel, WME & TKO", 2026-10-07. Stories we tell ourselves are not reality
- [Daron Acemoğlu: The money in AI is in data and applications, not models](https://realalpha.app/s/the-money-in-ai-is-in-data-and-applications-not-models/index.md): from "Nobel Economist: Where He Sees the Real Money in AI", 2026-10-06. Open source models limit how much foundation models can monetize
- [Daron Acemoğlu: Making workers 5% more productive beats automating 5% of jobs](https://realalpha.app/s/making-workers-5-more-productive-beats-automating-5-of-jobs/index.md): from "Nobel Economist: Where He Sees the Real Money in AI", 2026-10-06. Augmenting all workers can outperform automating a small share of jobs
- [Kevin Mandia: AI has to take over the detect-and-respond loop](https://realalpha.app/s/ai-has-to-take-over-the-detect-and-respond-loop/index.md): from "The Biggest Shift in Cybersecurity in 30 Years", 2026-10-06. Human response times will be too slow for machine-speed attacks.
- [Kevin Mandia: Cyber agents need domain expertise and layered safety checks](https://realalpha.app/s/cyber-agents-need-domain-expertise-and-layered-safety-checks/index.md): from "The Biggest Shift in Cybersecurity in 30 Years", 2026-10-06. Pair AI teams with security operators who understand real attack chains.
- [Kevin Mandia: AI makes attacks faster and could reshape nation-state tactics](https://realalpha.app/s/ai-makes-attacks-faster-and-could-reshape-nation-state/index.md): from "The Biggest Shift in Cybersecurity in 30 Years", 2026-10-06. AI can explore more attack paths than human teams can.
- [Kevin Mandia: Armadin maps networks and attacks when they change](https://realalpha.app/s/armadin-maps-networks-and-attacks-when-they-change/index.md): from "The Biggest Shift in Cybersecurity in 30 Years", 2026-10-06. A hyperattack builds an attacker’s view of the network.
- [Kevin Mandia: Mandiant built its model around breaches being inevitable](https://realalpha.app/s/mandiant-built-its-model-around-breaches-being-inevitable/index.md): from "The Biggest Shift in Cybersecurity in 30 Years", 2026-10-06. Mandiant’s first premise was that prevention alone would fail.
- [Yash Patil: Post-training wins inference because the biggest workloads train first](https://realalpha.app/s/post-training-wins-inference-because-the-biggest-workloads/index.md): from "Applied Compute CEO on the Limits of RL, the New AI Hyperscaler & Why Post-Training Wins Inference", 2026-10-06. Companies start on frontier models, then post-train once a workload matures
- [Yash Patil: RL is a hill-climbing machine, so protect your evals](https://realalpha.app/s/rl-is-a-hill-climbing-machine-so-protect-your-evals/index.md): from "Applied Compute CEO on the Limits of RL, the New AI Hyperscaler & Why Post-Training Wins Inference", 2026-10-06. Rubric-based RL against expert answers works well as a proxy
- [Yash Patil: Continual learning is blocked by data-efficient sparse-reward training](https://realalpha.app/s/continual-learning-is-blocked-by-data-efficient-sparse/index.md): from "Applied Compute CEO on the Limits of RL, the New AI Hyperscaler & Why Post-Training Wins Inference", 2026-10-06. Pre-training generalizes well but is not data efficient
- [Amin Vahdat: Goodput, not FLOPS, is how Google holds itself accountable](https://realalpha.app/s/goodput-not-flops-is-how-google-holds-itself-accountable/index.md): from "Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI", 2026-10-06. Synchronous workloads stall when one component fails, wasting redone work
- [Amin Vahdat: Google split the TPU into 8i and 8t based on market size](https://realalpha.app/s/google-split-the-tpu-into-8i-and-8t-based-on-market-size/index.md): from "Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI", 2026-10-06. Specialize only if the workload is big and durable, not a three-month spike
- [Amin Vahdat: Long-horizon agents push CPU, storage and network demand up](https://realalpha.app/s/long-horizon-agents-push-cpu-storage-and-network-demand-up/index.md): from "Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI", 2026-10-06. Without a human in the loop, request gaps fall from seconds to milliseconds
- [Ioannis Antonoglou: Beam targets strong reasoning with fewer tokens](https://realalpha.app/s/beam-targets-strong-reasoning-with-fewer-tokens/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. Beam has 500 billion total parameters and 23 billion active.
- [Misha Laskin: Open models can help make AI safer](https://realalpha.app/s/open-models-can-help-make-ai-safer/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. Open software has long relied on broad review to uncover vulnerabilities.
- [Misha Laskin: Open models can drive demand for a full stack](https://realalpha.app/s/open-models-can-drive-demand-for-a-full-stack/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. Open models can create demand for cloud and inference services.
- [Misha Laskin: Reflection expects larger models as reinforcement learning scales](https://realalpha.app/s/reflection-expects-larger-models-as-reinforcement-learning/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. Reflection plans significantly larger models in 2027.
- [Misha Laskin: AI provider revenue depends on compute, intelligence and trust](https://realalpha.app/s/ai-provider-revenue-depends-on-compute-intelligence-and/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. Owning more of the stack can accumulate margin over time.
- [Misha Laskin: Reflection began with bets on RL and Western open models](https://realalpha.app/s/reflection-began-with-bets-on-rl-and-western-open-models/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. The founders saw reinforcement learning as a path to agentic coding.
- [Misha Laskin: Data center owners need open models and a software partner](https://realalpha.app/s/data-center-owners-need-open-models-and-a-software-partner/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. Land, power and shell can bottleneck compute deployment.
- [Misha Laskin: Open models can fill gaps when closed safeguards block use](https://realalpha.app/s/open-models-can-fill-gaps-when-closed-safeguards-block-use/index.md): from "An open model challenger to Anthropic and OpenAI", 2026-10-05. They see release thresholds as capability-based, not openness-based.
- [Sam Parr: Movie studios spend more on marketing than on making the film](https://realalpha.app/s/movie-studios-spend-more-on-marketing-than-on-making-the/index.md): from "3 businesses we think are about to blow up", 2026-10-05. A $14M movie carried a $40M marketing budget, $54M all in
- [Jeanne Grosser: Never accept a founder's forecast without the pipeline math](https://realalpha.app/s/never-accept-a-founders-forecast-without-the-pipeline-math/index.md): from "How Vercel Is Rethinking Go-to-Market With AI | COO Jeanne Grosser", 2026-10-05. Founders are default optimists and rarely know how to forecast
- [Jeanne Grosser: Build an operating model to show what must be true](https://realalpha.app/s/build-an-operating-model-to-show-what-must-be-true/index.md): from "How Vercel Is Rethinking Go-to-Market With AI | COO Jeanne Grosser", 2026-10-05. Model funnel math by segment and geo, even with messy data
- [Jeanne Grosser: Vercel runs support, BDR and RevOps with agents](https://realalpha.app/s/vercel-runs-support-bdr-and-revops-with-agents/index.md): from "How Vercel Is Rethinking Go-to-Market With AI | COO Jeanne Grosser", 2026-10-05. Remaining support staff find product bugs and write PRs
- [Sam Stephenson: A design system helps engineers and agents ship consistently](https://realalpha.app/s/a-design-system-helps-engineers-and-agents-ship-consistently/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. Encode the design system in reusable components and tokens.
- [Sam Stephenson: An internal agent can become a tool everyone improves](https://realalpha.app/s/an-internal-agent-can-become-a-tool-everyone-improves/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. Connect an assistant to the internal sources people already use.
- [Sam Stephenson: Easy shipping makes product quality matter more](https://realalpha.app/s/easy-shipping-makes-product-quality-matter-more/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. The hard part is deciding what is worth building.
- [Sam Stephenson: Granola helps Sam prepare for and recover from meetings](https://realalpha.app/s/granola-helps-sam-prepare-for-and-recover-from-meetings/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. Candidate briefs help Sam avoid entering interviews unprepared.
- [Sam Stephenson: Granola hides the transcript to keep users present](https://realalpha.app/s/granola-hides-the-transcript-to-keep-users-present/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. Calm is the core product feeling during hectic transitions between meetings.
- [Sam Stephenson: Granola keeps private meetings separate from shared team notes](https://realalpha.app/s/granola-keeps-private-meetings-separate-from-shared-team/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. Managers cannot search employees’ private meeting notes.
- [Sam Stephenson: Granola lets enterprise workflows move beyond its UI](https://realalpha.app/s/granola-lets-enterprise-workflows-move-beyond-its-ui/index.md): from "How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson", 2026-10-04. Granola initially tried to keep users inside its own UI.
- [Tibo Sottiaux: Most internet actions will soon be taken by agents](https://realalpha.app/s/most-internet-actions-will-soon-be-taken-by-agents/index.md): from "OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux", 2026-10-04. Build as if models are 10 times better in a year
- [Tibo Sottiaux: OpenAI will rev-share with plugins and recommend by retention](https://realalpha.app/s/openai-will-rev-share-with-plugins-and-recommend-by/index.md): from "OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux", 2026-10-04. Sign in with ChatGPT launched with 16 partners
- [Tibo Sottiaux: A Dot warned Tibo of a production outage before a live demo](https://realalpha.app/s/a-dot-warned-tibo-of-a-production-outage-before-a-live-demo/index.md): from "OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux", 2026-10-04. The agent connected the outage to the upcoming live demo
- [David Friedberg: Cyber defense will force governments to allocate GPUs](https://realalpha.app/s/cyber-defense-will-force-governments-to-allocate-gpus/index.md): from "Trump’s Super Intelligence Summit, AI Safety Accord, GDP Beats, Midterm Predictions", 2026-10-02. Open-weight models make regulating the software impractical
- [David Sacks: The White House AI accord gets its teeth from board liability](https://realalpha.app/s/the-white-house-ai-accord-gets-its-teeth-from-board/index.md): from "Trump’s Super Intelligence Summit, AI Safety Accord, GDP Beats, Midterm Predictions", 2026-10-02. External auditors report to an independent board committee
- [David Sacks: Strong economic data contradicts the gloomy narrative](https://realalpha.app/s/strong-economic-data-contradicts-the-gloomy-narrative/index.md): from "Trump’s Super Intelligence Summit, AI Safety Accord, GDP Beats, Midterm Predictions", 2026-10-02. Q2 GDP was revised up to 2.2% from 1.5%
- [Alex Zhang: RLMs generalize across tasks because the strategy stays the same](https://realalpha.app/s/rlms-generalize-across-tasks-because-the-strategy-stays-the/index.md): from "Recursive Language Models — Alex Zhang, MIT PhD", 2026-10-02. Offloading context to code makes different tasks share one top-level strategy
- [Alex Zhang: Academics should take big bets that industry labs won't](https://realalpha.app/s/academics-should-take-big-bets-that-industry-labs-wont/index.md): from "Recursive Language Models — Alex Zhang, MIT PhD", 2026-10-02. Dismissive reactions to a simple idea can signal nobody sees its purpose yet
- [Alex Zhang: One expert insight can erase a trillion tokens of search](https://realalpha.app/s/one-expert-insight-can-erase-a-trillion-tokens-of-search/index.md): from "Recursive Language Models — Alex Zhang, MIT PhD", 2026-10-02. Domain experts still act as strong verifiers and steerers of AI
- [Si Sheppard: The natives adapted fast, but the learning curve was too steep](https://realalpha.app/s/the-natives-adapted-fast-but-the-learning-curve-was-too/index.md): from "How did a few hundred Spanish soldiers topple two empires? – Si Sheppard", 2026-10-01. Aztecs drew the Spanish into a city of causeways, which blunted the cavalry
- [Si Sheppard: The Tlaxcalans used Cortés, not the other way around](https://realalpha.app/s/the-tlaxcalans-used-cortes-not-the-other-way-around/index.md): from "How did a few hundred Spanish soldiers topple two empires? – Si Sheppard", 2026-10-01. Tlaxcalans demanded territory and a fortress in Tenochtitlan as their price
- [Sarah Wang: AI adoption looks broad but is shallow, with power users far ahead](https://realalpha.app/s/ai-adoption-looks-broad-but-is-shallow-with-power-users-far/index.md): from "AI, Infrastructure, and the Next Investment Cycle", 2026-09-30. 69% of S&P 500 firms have live AI deployments, but only 2% track results over time
- [Alex Immerman: Agents may cost marketplaces ad revenue but could lift total orders](https://realalpha.app/s/agents-may-cost-marketplaces-ad-revenue-but-could-lift-total/index.md): from "AI, Infrastructure, and the Next Investment Cycle", 2026-09-30. Ask how much incremental demand an agent brings and how much profit comes from discovery
- [Bill Ackman: A venture reset is coming, so founders should hoard runway](https://realalpha.app/s/a-venture-reset-is-coming-so-founders-should-hoard-runway/index.md): from "Bill Ackman: People are Going to Lose a Lot of Money", 2026-09-29. Valuations jumped from $400M to $1B in two weeks in one deal he saw
- [Thariq Shihipar: Agents hacked their scorer and chained exploits to communicate](https://realalpha.app/s/agents-hacked-their-scorer-and-chained-exploits-to/index.md): from "Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic", 2026-09-29. Agents with fixed compute budgets started collaborating to finish an impossible task
- [Thariq Shihipar: Match model effort to the task and ask for decision notes](https://realalpha.app/s/match-model-effort-to-the-task-and-ask-for-decision-notes/index.md): from "Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic", 2026-09-29. Use high or max effort for code review and security, low for UI work
- [Thariq Shihipar: Prompting is a high-skill craft built on a mental model of Claude](https://realalpha.app/s/prompting-is-a-high-skill-craft-built-on-a-mental-model-of/index.md): from "Claude Code’s Bitter Lesson: Prompts, Harnesses, Mods, and Agents — Thariq Shihipar, Anthropic", 2026-09-29. Build a mental model of what Claude can one-shot and what it can't
- [Lauren Tan: Build trust in a bot by turning observed runs into skills](https://realalpha.app/s/build-trust-in-a-bot-by-turning-observed-runs-into-skills/index.md): from "We Built Grok Bot. Here Are Our 14 Best Bots | Peng Zheng & Lauren Tan", 2026-09-27. Observe the bot doing the task and course correct it
- [Peng Zheng, Lauren Tan: An eng lead bot that only delegates lets one person run agent swarms](https://realalpha.app/s/an-eng-lead-bot-that-only-delegates-lets-one-person-run/index.md): from "We Built Grok Bot. Here Are Our 14 Best Bots | Peng Zheng & Lauren Tan", 2026-09-27. Tell the eng lead bot to never work itself, only break down and delegate
- [Peng Zheng: Persistent named bots beat disposable chat sessions](https://realalpha.app/s/persistent-named-bots-beat-disposable-chat-sessions/index.md): from "We Built Grok Bot. Here Are Our 14 Best Bots | Peng Zheng & Lauren Tan", 2026-09-27. Group chats let bots with different roles debate an idea
- [Garry Tan: Solo founders hit 19% of YC as knowing what to build matters most](https://realalpha.app/s/solo-founders-hit-19-of-yc-as-knowing-what-to-build-matters/index.md): from "The State of Startups in 2026", 2026-09-17. Solo founders rose from about 5% to over 18-19% of accepted companies
- [Jared Friedman: RL environment and data startups quietly build huge revenue](https://realalpha.app/s/rl-environment-and-data-startups-quietly-build-huge-revenue/index.md): from "The State of Startups in 2026", 2026-09-17. More than a dozen YC companies each earn over $10M a year selling data to labs
- [Ali Ghodsi: Great enterprise sellers are aggressive, high-EQ and not technical](https://realalpha.app/s/great-enterprise-sellers-are-aggressive-high-eq-and-not/index.md): from "Databricks’ Ali Ghodsi Never Wanted to Be CEO. Now He’s Among the Best", 2026-09-17. Technical depth did not predict top sales performance
- [Ali Ghodsi: Hire executives early and be extremely picky to avoid false positives](https://realalpha.app/s/hire-executives-early-and-be-extremely-picky-to-avoid-false/index.md): from "Databricks’ Ali Ghodsi Never Wanted to Be CEO. Now He’s Among the Best", 2026-09-17. Start exec searches before you need the person, six to twelve months
- [Ali Ghodsi: Pick one giant bottleneck and give it years of extreme focus](https://realalpha.app/s/pick-one-giant-bottleneck-and-give-it-years-of-extreme-focus/index.md): from "Databricks’ Ali Ghodsi Never Wanted to Be CEO. Now He’s Among the Best", 2026-09-17. Identify the single biggest bottleneck and overweight it
- [Andrew Scull: Antidepressants beat placebo but often not by a meaningful margin](https://realalpha.app/s/antidepressants-beat-placebo-but-often-not-by-a-meaningful/index.md): from "Psychiatry, Insane Asylums, Mental Illness, ECT, Lobotomies, Freud & Jung | Lex Fridman Podcast #502", 2026-09-17. A one or two point gain on a 60-point scale can still be statistically significant
- [Tobi Lütke: Intuition matters most when the best path has no feedback](https://realalpha.app/s/intuition-matters-most-when-the-best-path-has-no-feedback/index.md): from "Tobi Lütke: The Skills AI Makes More Valuable", 2026-09-15. Intuition is most valuable where no feedback mechanism exists
- [Tobi Lütke: The hard part is choosing among many good options](https://realalpha.app/s/the-hard-part-is-choosing-among-many-good-options/index.md): from "Tobi Lütke: The Skills AI Makes More Valuable", 2026-09-15. Prune irrelevant options first, then judge the ones that remain
- [Tobi Lütke: Machines can't take responsibility, so use an AI council](https://realalpha.app/s/machines-cant-take-responsibility-so-use-an-ai-council/index.md): from "Tobi Lütke: The Skills AI Makes More Valuable", 2026-09-15. Humans must own the decision; AI informs it
- [Buck Shlegeris: Why AIs that only slightly like each other can still collude](https://realalpha.app/s/why-ais-that-only-slightly-like-each-other-can-still-collude/index.md): from "Top Alignment Researcher on OpenAI/HuggingFace Revelations, Fixing AI Safety & Takeover Odds", 2026-09-03. Paperclip-style AIs share a goal, so a monitor would let a bad action slide
- [Buck Shlegeris: Sloppy RL environments push models to obsess over graders](https://realalpha.app/s/sloppy-rl-environments-push-models-to-obsess-over-graders/index.md): from "Top Alignment Researcher on OpenAI/HuggingFace Revelations, Fixing AI Safety & Takeover Odds", 2026-09-03. Sloppy tasks reward guessing the grader instead of taking instructions at face value
- [David Heinemeier Hansson: Why big companies don't ship faster with AI agents](https://realalpha.app/s/why-big-companies-dont-ship-faster-with-ai-agents/index.md): from "DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux | Lex Fridman Podcast #501", 2026-08-26. Human communication layers, not coding, kill productivity in teams
- [Khabib Nurmagomedov: Why Khabib's father had him compete in every combat sport](https://realalpha.app/s/why-khabibs-father-had-him-compete-in-every-combat-sport/index.md): from "Khabib Nurmagomedov: Dagestan, MMA, UFC, Islam, Conor, Fedor & Football | Lex Fridman Podcast #500", 2026-08-12. A single triangle choke convinced him wrestling alone was not enough
- [Khabib Nurmagomedov: Staying a champion is harder than becoming one](https://realalpha.app/s/staying-a-champion-is-harder-than-becoming-one/index.md): from "Khabib Nurmagomedov: Dagestan, MMA, UFC, Islam, Conor, Fedor & Football | Lex Fridman Podcast #500", 2026-08-12. Distractions cause missed workouts, which then pile up
- [Brad Jacobs: Improvising on a pivot: how a messy org chart became a 15x deal](https://realalpha.app/s/improvising-on-a-pivot-how-a-messy-org-chart-became-a-15x/index.md): from "The Business Lessons That Made Him Billions", 2026-08-04. A messy, duplicated org chart is a sign of easy money to be made
- [Brad Jacobs: How to pick an industry and price an acquisition](https://realalpha.app/s/how-to-pick-an-industry-and-price-an-acquisition/index.md): from "The Business Lessons That Made Him Billions", 2026-08-04. Select an industry first: big, fragmented, with scale advantages
- [Brad Jacobs: The diligence questions Jacobs asks the top 15 people](https://realalpha.app/s/the-diligence-questions-jacobs-asks-the-top-15-people/index.md): from "The Business Lessons That Made Him Billions", 2026-08-04. Ask: if this was your money, would you buy this company?
- [Gary Gallagher: The Seven Days, not Antietam, pushed Lincoln to emancipation](https://realalpha.app/s/the-seven-days-not-antietam-pushed-lincoln-to-emancipation/index.md): from "Gary Gallagher: American Civil War, Slavery, Lincoln, Grant & Lee | Lex Fridman Podcast #499", 2026-07-28. By May 1862 the Confederacy had lost New Orleans, Nashville and Memphis
- [Gary Gallagher: Why Union morale hit bottom in summer 1864](https://realalpha.app/s/why-union-morale-hit-bottom-in-summer-1864/index.md): from "Gary Gallagher: American Civil War, Slavery, Lincoln, Grant & Lee | Lex Fridman Podcast #499", 2026-07-28. Grant fought Lee continuously, with no pauses between battles
- [Gary Gallagher: The Confederacy only had to outlast Northern morale](https://realalpha.app/s/the-confederacy-only-had-to-outlast-northern-morale/index.md): from "Gary Gallagher: American Civil War, Slavery, Lincoln, Grant & Lee | Lex Fridman Podcast #499", 2026-07-28. The Confederacy never had to invade or capture Washington
- [Gio Valiante: Climb out of a drawdown by making small money again](https://realalpha.app/s/climb-out-of-a-drawdown-by-making-small-money-again/index.md): from "No.1 Performance Psychologist: The Secret to High Performance and Excellence", 2026-07-07. Trying to recoup a drawdown in one swing triggers panic and bad decisions
- [Gio Valiante: Confidence comes from four sources, and failure outweighs success](https://realalpha.app/s/confidence-comes-from-four-sources-and-failure-outweighs/index.md): from "No.1 Performance Psychologist: The Secret to High Performance and Excellence", 2026-07-07. Ask why you believe you are good at something; the answer maps to four sources
- [Gio Valiante: Adult drive often comes from undoing childhood debris](https://realalpha.app/s/adult-drive-often-comes-from-undoing-childhood-debris/index.md): from "No.1 Performance Psychologist: The Secret to High Performance and Excellence", 2026-07-07. Deep motivation often traces back to formative youth experiences
- [Anthony Kaldellis: Emperors stayed honest because nearly half were overthrown](https://realalpha.app/s/emperors-stayed-honest-because-nearly-half-were-overthrown/index.md): from "The Rise and Fall of the Roman Empire and the Byzantine Empire | Lex Fridman Podcast #498", 2026-06-30. With no hereditary right to rule, every emperor was vulnerable
- [Anthony Kaldellis: The empire fell to invasions, not internal decay](https://realalpha.app/s/the-empire-fell-to-invasions-not-internal-decay/index.md): from "The Rise and Fall of the Roman Empire and the Byzantine Empire | Lex Fridman Podcast #498", 2026-06-30. No provincial separatism or peasant uprisings ever broke the tax system
- [Andrew Fox: Orbital compute could cut a gigawatt's cost from $60B to $30B](https://realalpha.app/s/orbital-compute-could-cut-a-gigawatts-cost-from-60b-to-30b/index.md): from "The SpaceX IPO, Fable 5, AI Capex Update & Market Check w/ Gavin Baker, Andrew Fox & Clark Tang", 2026-06-11. Launch cost could fall from about $1,500 per kg on Falcon to under $250 per kg
- [Gavin Baker: Frontier labs capture the revenue, but open source may carry most tokens](https://realalpha.app/s/frontier-labs-capture-the-revenue-but-open-source-may-carry/index.md): from "The SpaceX IPO, Fable 5, AI Capex Update & Market Check w/ Gavin Baker, Andrew Fox & Clark Tang", 2026-06-11. Frontier may hold 90% of economic value while open source carries 80% of tokens
- [Brad Gerstner: Why $1.5T of AI CapEx can still pencil out against $300B of revenue](https://realalpha.app/s/why-1-5t-of-ai-capex-can-still-pencil-out-against-300b-of/index.md): from "The SpaceX IPO, Fable 5, AI Capex Update & Market Check w/ Gavin Baker, Andrew Fox & Clark Tang", 2026-06-11. Baker thinks the $300B revenue figure is low and 2026 will end well above $200B
- [Don Lincoln: Why a theory of everything is probably far off and string theory likely wrong](https://realalpha.app/s/why-a-theory-of-everything-is-probably-far-off-and-string/index.md): from "Biggest Mysteries in Physics: Antimatter, Dark Energy & ToE - Don Lincoln | Lex Fridman Podcast #497", 2026-05-29. Planck scale is about 10^15 times beyond today's accelerators
- [Don Lincoln: Making one gram of antimatter would take a billion years](https://realalpha.app/s/making-one-gram-of-antimatter-would-take-a-billion-years/index.md): from "Biggest Mysteries in Physics: Antimatter, Dark Energy & ToE - Don Lincoln | Lex Fridman Podcast #497", 2026-05-29. Fermilab needed 100,000 protons to make one antiproton
- [Don Lincoln: Why Fermilab narrowly missed the Higgs discovery to CERN](https://realalpha.app/s/why-fermilab-narrowly-missed-the-higgs-discovery-to-cern/index.md): from "Biggest Mysteries in Physics: Antimatter, Dark Energy & ToE - Don Lincoln | Lex Fridman Podcast #497", 2026-05-29. Fermilab needed two or three more years of data to find the Higgs
- [David Rosenthal: Bogle's firing and the legal loophole that created Vanguard](https://realalpha.app/s/bogles-firing-and-the-legal-loophole-that-created-vanguard/index.md): from "Vanguard: The communist capitalist who saved investors a trillion dollars (Audio)", 2026-05-18. Bogle proposed mutualizing the funds when nobody in the industry was asking for it
- [Ben Gilbert: A 1% fee quietly costs you a third of your retirement](https://realalpha.app/s/a-1-fee-quietly-costs-you-a-third-of-your-retirement/index.md): from "Vanguard: The communist capitalist who saved investors a trillion dollars (Audio)", 2026-05-18. A 1% fee on 7% market returns takes about 15% of your gains
- [David Rosenthal: Buffett's million-dollar bet: an index fund crushed hedge funds](https://realalpha.app/s/buffetts-million-dollar-bet-an-index-fund-crushed-hedge/index.md): from "Vanguard: The communist capitalist who saved investors a trillion dollars (Audio)", 2026-05-18. Only one hedge fund manager, Ted Seides, accepted Buffett's bet
- [Yann LeCun: A world model predicts consequences, which LLMs cannot do](https://realalpha.app/s/a-world-model-predicts-consequences-which-llms-cannot-do/index.md): from "Yann LeCun on What Comes After LLMs", 2026-05-15. Agents need to predict the consequences of their own actions
- [Yann LeCun: Tapestry would train an open sovereign model without sharing data](https://realalpha.app/s/tapestry-would-train-an-open-sovereign-model-without-sharing/index.md): from "Yann LeCun on What Comes After LLMs", 2026-05-15. Contributors keep their data and share only parameter vectors
- [Yann LeCun: LLMs excel only where language is the substrate of reasoning](https://realalpha.app/s/llms-excel-only-where-language-is-the-substrate-of-reasoning/index.md): from "Yann LeCun on What Comes After LLMs", 2026-05-15. Math and code are domains where language is the reasoning substrate
- [Jean-Baptiste Kempf: Why hand-written assembly beats compilers by 10x to 50x](https://realalpha.app/s/why-hand-written-assembly-beats-compilers-by-10x-to-50x/index.md): from "FFmpeg: The Incredible Technology Behind Video on the Internet | Lex Fridman Podcast #496", 2026-05-06. Hand-written SIMD gives 10x to 50x gains over C; auto-vectorization is not close
- [Jean-Baptiste Kempf: Why Kempf turned down tens of millions to keep VLC ad-free](https://realalpha.app/s/why-kempf-turned-down-tens-of-millions-to-keep-vlc-ad-free/index.md): from "FFmpeg: The Incredible Technology Behind Video on the Internet | Lex Fridman Podcast #496", 2026-05-06. Only shady ad and toolbar companies ever offered money
- [Kieran Kunhya: How Google's AI bug reports clashed with FFmpeg volunteers](https://realalpha.app/s/how-googles-ai-bug-reports-clashed-with-ffmpeg-volunteers/index.md): from "FFmpeg: The Incredible Technology Behind Video on the Internet | Lex Fridman Podcast #496", 2026-05-06. AI reports flooded volunteers on niche codecs, all marked high priority
- [Lex Fridman: Eric the Red's Greenland pitch was history's best real estate scam](https://realalpha.app/s/eric-the-reds-greenland-pitch-was-historys-best-real-estate/index.md): from "Vikings, Ragnar, Berserkers, Valhalla & the Warriors of the Viking Age | Lex Fridman Podcast #495", 2026-04-09. Eric was exiled from Norway, then Iceland, and chose to sail west
- [Lars Brownworth: Vikings used terror and inside knowledge to pick targets](https://realalpha.app/s/vikings-used-terror-and-inside-knowledge-to-pick-targets/index.md): from "Vikings, Ragnar, Berserkers, Valhalla & the Warriors of the Viking Age | Lex Fridman Podcast #495", 2026-04-09. They struck on Easter and Christmas, when churches held the most wealth
- [Lars Brownworth: Rollo refused to kiss the king's foot, so a guard did it](https://realalpha.app/s/rollo-refused-to-kiss-the-kings-foot-so-a-guard-did-it/index.md): from "Vikings, Ragnar, Berserkers, Valhalla & the Warriors of the Viking Age | Lex Fridman Podcast #495", 2026-04-09. Rollo's treaty gave him an ambiguous title, which his heirs exploited
- [Jakub Pachocki: Why OpenAI hides chains of thought to keep them monitorable](https://realalpha.app/s/why-openai-hides-chains-of-thought-to-keep-them-monitorable/index.md): from "OpenAI’s Chief Scientist on Continual Learning Hype, RL Beyond Code, & Future Alignment Directions", 2026-04-09. Unsupervised chains of thought stay honest and are written in English
- [Jakub Pachocki: In-context learning may beat custom RL for most companies](https://realalpha.app/s/in-context-learning-may-beat-custom-rl-for-most-companies/index.md): from "OpenAI’s Chief Scientist on Continual Learning Hype, RL Beyond Code, & Future Alignment Directions", 2026-04-09. Still build evals and gather examples for your domain
- [Jakub Pachocki: Continual learning is what scaling GPT models was always for](https://realalpha.app/s/continual-learning-is-what-scaling-gpt-models-was-always-for/index.md): from "OpenAI’s Chief Scientist on Continual Learning Hype, RL Beyond Code, & Future Alignment Directions", 2026-04-09. GPT-3 was framed around learning to learn in context
- [Jensen Huang: Putting CUDA on GeForce nearly sank NVIDIA, but install base won](https://realalpha.app/s/putting-cuda-on-geforce-nearly-sank-nvidia-but-install-base/index.md): from "Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494", 2026-03-23. Install base defines a computing architecture, more than elegance does
- [Jensen Huang: Test every plan against the speed of light, not continuous improvement](https://realalpha.app/s/test-every-plan-against-the-speed-of-light-not-continuous/index.md): from "Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494", 2026-03-23. Set the physical limit first, then explain the gap to today's process
- [Jensen Huang: The power grid sits mostly idle, and data centers could use that slack](https://realalpha.app/s/the-power-grid-sits-mostly-idle-and-data-centers-could-use/index.md): from "Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution | Lex Fridman Podcast #494", 2026-03-23. Grids run around 60% of peak 99% of the time
- [Alex Karp: Military technology helps deter attacks and bring warfighters home](https://realalpha.app/s/military-technology-helps-deter-attacks-and-bring/index.md): from "Palantir CEO on Iran, AI Weapons and American Domination | a16z American Dynamism Summit", 2026-03-12. Karp says technology is a key part of modern warfighting.
- [Alex Karp: AI firms risk nationalization if they ignore public concerns](https://realalpha.app/s/ai-firms-risk-nationalization-if-they-ignore-public-concerns/index.md): from "Palantir CEO on Iran, AI Weapons and American Domination | a16z American Dynamism Summit", 2026-03-12. Job losses and defense concerns could unite political opponents.
- [Alex Karp: America’s edge is cultivating unconventional talent](https://realalpha.app/s/americas-edge-is-cultivating-unconventional-talent/index.md): from "Palantir CEO on Iran, AI Weapons and American Domination | a16z American Dynamism Summit", 2026-03-12. Karp values people for their distinctive abilities, not their politics.
- [Alex Karp: AI firms need a shared plan for privacy, jobs and battlefield use](https://realalpha.app/s/ai-firms-need-a-shared-plan-for-privacy-jobs-and-battlefield/index.md): from "Palantir CEO on Iran, AI Weapons and American Domination | a16z American Dynamism Summit", 2026-03-12. Organize early, as Hollywood did with its rating system.
- [Alex Karp: Defense founders should meet warfighters and know their limits](https://realalpha.app/s/defense-founders-should-meet-warfighters-and-know-their/index.md): from "Palantir CEO on Iran, AI Weapons and American Domination | a16z American Dynamism Summit", 2026-03-12. Visit a base or Iowa before making defense decisions.
- [Jeff Kaplan: Why Titan failed: hiring ahead of a vision](https://realalpha.app/s/why-titan-failed-hiring-ahead-of-a-vision/index.md): from "Jeff Kaplan: World of Warcraft, Overwatch, Blizzard, and Future of Gaming | Lex Fridman Podcast #493", 2026-03-11. Ideas are cheap; vision needs a tech, design, art and production plan
- [Jeff Kaplan: How WoW's quest design made leveling a path of least resistance](https://realalpha.app/s/how-wows-quest-design-made-leveling-a-path-of-least/index.md): from "Jeff Kaplan: World of Warcraft, Overwatch, Blizzard, and Future of Gaming | Lex Fridman Podcast #493", 2026-03-11. Testers with no EverQuest background exposed that the quest count was far too low
- [Jeff Kaplan: What broke Kaplan at Blizzard: revenue targets and Overwatch 2](https://realalpha.app/s/what-broke-kaplan-at-blizzard-revenue-targets-and-overwatch/index.md): from "Jeff Kaplan: World of Warcraft, Overwatch, Blizzard, and Future of Gaming | Lex Fridman Podcast #493", 2026-03-11. Overwatch League deals forced in-game commitments that derailed the roadmap
- [Rick Beato: Why AI songs get spotted, and who benefits from AI](https://realalpha.app/s/why-ai-songs-get-spotted-and-who-benefits-from-ai/index.md): from "Rick Beato: Greatest Guitarists of All Time, History & Future of Music | Lex Fridman Podcast #492", 2026-03-01. Lyrics from Claude, music from Suno; about three good songs per 130 ideas
- [Rick Beato: Fighting YouTube copyright claims on fair use paid off](https://realalpha.app/s/fighting-youtube-copyright-claims-on-fair-use-paid-off/index.md): from "Rick Beato: Greatest Guitarists of All Time, History & Future of Music | Lex Fridman Podcast #492", 2026-03-01. Labels claim all revenue even for 20 seconds of a clip
- [Rick Beato: Why the Beatles' best work came once they stopped touring](https://realalpha.app/s/why-the-beatles-best-work-came-once-they-stopped-touring/index.md): from "Rick Beato: Greatest Guitarists of All Time, History & Future of Music | Lex Fridman Podcast #492", 2026-03-01. Bad monitoring and screaming fans ended touring
- [Peter Steinberger: Squatters sniped his accounts in seconds during the rename](https://realalpha.app/s/squatters-sniped-his-accounts-in-seconds-during-the-rename/index.md): from "OpenClaw: The Viral AI Agent that Broke the Internet - Peter Steinberger | Lex Fridman Podcast #491", 2026-02-12. No platform offers squatter protection during a rename
- [Peter Steinberger: Empathize with the agent's blank-slate view of your code](https://realalpha.app/s/empathize-with-the-agents-blank-slate-view-of-your-code/index.md): from "OpenClaw: The Viral AI Agent that Broke the Internet - Peter Steinberger | Lex Fridman Podcast #491", 2026-02-12. Agents start fresh each session, so point them where to look
- [Alex Karp: Palantir builds for what clients need, not what traps them](https://realalpha.app/s/palantir-builds-for-what-clients-need-not-what-traps-them/index.md): from "Alex Karp, CEO of Palantir: Exclusive Interview Inside PLTR Office", 2025-11-11. Build what customers ought to need, not just what they ask for.
- [Alex Karp: AI buyers now seek working systems, not proof that software works](https://realalpha.app/s/ai-buyers-now-seek-working-systems-not-proof-that-software/index.md): from "Alex Karp, CEO of Palantir: Exclusive Interview Inside PLTR Office", 2025-11-11. Buyers increasingly arrive convinced that AI software can work.
- [Alex Karp: Palantir uses software to help US workers scale factory know-how](https://realalpha.app/s/palantir-uses-software-to-help-us-workers-scale-factory-know/index.md): from "Alex Karp, CEO of Palantir: Exclusive Interview Inside PLTR Office", 2025-11-11. He says US plants need to adopt the processes of the factories they emulate.
- [Alex Karp: Karp sees deterrence as a way to keep US troops from fighting](https://realalpha.app/s/karp-sees-deterrence-as-a-way-to-keep-us-troops-from/index.md): from "Alex Karp, CEO of Palantir: Exclusive Interview Inside PLTR Office", 2025-11-11. Karp says his time in Germany shaped how he understands the Nazi era.
- [Alex Karp: Karp says his outsider background shaped Palantir’s worker focus](https://realalpha.app/s/karp-says-his-outsider-background-shaped-palantirs-worker/index.md): from "Alex Karp, CEO of Palantir: Exclusive Interview Inside PLTR Office", 2025-11-11. He sees his outsider background as a bridge to working-class Americans.
- [Satya Nadella: Compute demand depends on price, and a glut will come](https://realalpha.app/s/compute-demand-depends-on-price-and-a-glut-will-come/index.md): from "All things AI w @altcap @sama & @satyanadella. A Halloween Special. 🎃🔥BG2 w/ Brad Gerstner", 2025-10-31. Demand for compute only makes sense at a given price per unit of intelligence
- [Satya Nadella: Why Microsoft built a fungible fleet and said no to OpenAI](https://realalpha.app/s/why-microsoft-built-a-fungible-fleet-and-said-no-to-openai/index.md): from "All things AI w @altcap @sama & @satyanadella. A Halloween Special. 🎃🔥BG2 w/ Brad Gerstner", 2025-10-31. Fungibility across workloads, regions and chip generations drove the build plan
- [Satya Nadella: SaaS will survive AI if it is low ARPU and high usage](https://realalpha.app/s/saas-will-survive-ai-if-it-is-low-arpu-and-high-usage/index.md): from "All things AI w @altcap @sama & @satyanadella. A Halloween Special. 🎃🔥BG2 w/ Brad Gerstner", 2025-10-31. Agents replace the business logic tier, decoupling data, logic and UI
- [Jensen Huang: Why even free ASICs lose to NVIDIA when power is the limit](https://realalpha.app/s/why-even-free-asics-lose-to-nvidia-when-power-is-the-limit/index.md): from "NVIDIA: OpenAI, Future of Compute, and the American Dream | BG2 w/ Bill Gurley and Brad Gerstner", 2025-09-26. Power-limited buyers should maximize revenue per watt, not chip price
- [Jensen Huang: The math behind a $5 trillion annual AI infrastructure market](https://realalpha.app/s/the-math-behind-a-5-trillion-annual-ai-infrastructure-market/index.md): from "NVIDIA: OpenAI, Future of Compute, and the American Dream | BG2 w/ Bill Gurley and Brad Gerstner", 2025-09-26. Compute refresh moves from general purpose CPUs to accelerated computing
- [Alex Karp: Palantir says its safeguards make data analytics harder to abuse](https://realalpha.app/s/palantir-says-its-safeguards-make-data-analytics-harder-to/index.md): from "Palantir CEO Alex Karp: Why the West is Destroying Itself, Data Empire, Skeptics, How to Win", 2025-09-09. Access controls limit who can see and use data.
- [Alex Karp: The West weakens when it stops valuing its own culture](https://realalpha.app/s/the-west-weakens-when-it-stops-valuing-its-own-culture/index.md): from "Palantir CEO Alex Karp: Why the West is Destroying Itself, Data Empire, Skeptics, How to Win", 2025-09-09. Karp points to Germany’s lost industrial and energy strengths.
- [Alex Karp: China is best countered by strengthening the United States](https://realalpha.app/s/china-is-best-countered-by-strengthening-the-united-states/index.md): from "Palantir CEO Alex Karp: Why the West is Destroying Itself, Data Empire, Skeptics, How to Win", 2025-09-09. Treat domestic stability as a strategic priority.
- [Alex Karp: Karp links open borders to lower worker value](https://realalpha.app/s/karp-links-open-borders-to-lower-worker-value/index.md): from "Palantir CEO Alex Karp: Why the West is Destroying Itself, Data Empire, Skeptics, How to Win", 2025-09-09. He links open borders to political avoidance of workers’ economic problems.
- [Bill Gurley: Xiaomi built a car factory in three years with 2,000 workers](https://realalpha.app/s/xiaomi-built-a-car-factory-in-three-years-with-2-000-workers/index.md): from "China, China, China. Breaking Down China’s Tech Surge | BG2 w/ Bill Gurley and Brad Gerstner", 2025-08-28. Xiaomi makes 1,000 cars a day with 2,000 employees and has a 30 to 40 week backlog
- [The subsidy and theft story about China doesn't explain its success](https://realalpha.app/s/the-subsidy-and-theft-story-about-china-doesnt-explain-its/index.md): from "China, China, China. Breaking Down China’s Tech Surge | BG2 w/ Bill Gurley and Brad Gerstner", 2025-08-28. BYD says it receives no meaningful government subsidy
- [Bill Gurley: Why Western venture investment in China has dried up](https://realalpha.app/s/why-western-venture-investment-in-china-has-dried-up/index.md): from "China, China, China. Breaking Down China’s Tech Surge | BG2 w/ Bill Gurley and Brad Gerstner", 2025-08-28. Sequoia and GGV split, so far fewer Western dollars are available
- [Philippe Laffont: AI productivity could bend the US debt-to-GDP curve](https://realalpha.app/s/ai-productivity-could-bend-the-us-debt-to-gdp-curve/index.md): from "Coatue’s Laffont Brothers. AI, Public & VC Mkts, Macro, US Debt, Crypto, IPO's, & more | BG2", 2025-06-20. Markets were expensive in the 90s too, and still did well
- [Thomas Laffont: Headcount is flat while revenue grows at Mag7 and AppLovin](https://realalpha.app/s/headcount-is-flat-while-revenue-grows-at-mag7-and-applovin/index.md): from "Coatue’s Laffont Brothers. AI, Public & VC Mkts, Macro, US Debt, Crypto, IPO's, & more | BG2", 2025-06-20. Mag7 revenue grew over 20% a year while opex grew about 2%
- [Philippe Laffont: ChatGPT subscribers cut their Google page views by 8 to 11%](https://realalpha.app/s/chatgpt-subscribers-cut-their-google-page-views-by-8-to-11/index.md): from "Coatue’s Laffont Brothers. AI, Public & VC Mkts, Macro, US Debt, Crypto, IPO's, & more | BG2", 2025-06-20. Joining separate datasets is where the insight comes from