Alex Atallah (OpenRouter, OpenSea): Selling to Stripe for $7B+ and the Unfolding AI Economy

Our belief is that the revealed preferences that developers and businesses and enterprises have for language models and other AI models are really valuable and really important to have in one place on the internet. Models when we got started didn't have a place on the internet where you could see them all and see how people were using them in the real world and so that's how open router got started and as we've grown and we had like a 19 trillion token day in August which to put in perspective is about like one English Wikipedia every 30 seconds and and we grew about 50% month over month in terms of tokens this past month as we've grown the data becomes like more useful and we can provide pure benchmarks and help people discover the best models for their use cases we also help enterprises monitor observe see how everything is going on within their borders and govern it you have to put limits on how much people can spend and different team and which models different teams can access.

And then we help everybody. Get access to the best prices and best performance for each model that they end up caring about. What mindset were you in when you were going into building OpenRouter? Did you feel like you had something to prove or did you just feel like excited? You had like already just built something amazing. Where was your head at? I really wanted to start something new. I was kind of like feeling the itch to like, you know, have a small team in a new domain. and um and that was mostly where my head was at honestly like I um you know like OpenSea like this was before FTX um when I left um I stayed on the board um I I was you know when I wanted to like take a few months off or something but I think in like November ChatGPT launched it's about like two months after i left open sea and i was like oh sabbatical is probably over this is a crazy new thing um and and my biggest question was like is this going to be a winner

Alex Atallahtake all market this because this could be like a google style monopoly where all value flows to one company and it would be like the dutch east india trading company but like times you know a

quadrillion in magnitude if all knowledge work ends up depending on one company. That seems very dangerous and very, you know, I was also curious, like, who's using the other model labs? Like, who's using anything other than OpenAI? What are they using them for? Like nobody seems to have any empirical data on any of this stuff. and so I made like an experiment called window AI that was like kind of crypto inspired kind of like meta it was similar to MetaMask but where you'd bring your own users could like choose the model they want to use with a web app and bring their model choice directly to that that app and it allowed you know apps to kind of emerge that wanted to be model agnostic or wanted to let users use open weight models with them alpaca had had just come out which was the the first like usable open weight model i'd ever seen it was made and it only took six hundred dollars to make it it was a like a small team at stanford that um that set it up uh and it and like just fine-tuned llama using a bunch of synthetic data and i was like wow if this is possible then there's gonna be tons of these things. Maybe like thousands or tens of thousands of bottles. And it's a new way of selling data. It's a new way of kind of, of, of like monetizing knowledge. And also creating

like, yeah, it's, it's a new primitive building block for the economy. And there's, is there like a homepage? Is there any sort of like way to discover these things or any sort of place on the you know internet for them is there any sort of a way to get like high quality access to them is there a way to like compare you know the various providers that emerge over time is there a way to like monitor or you know manage observe and govern all of these things no so that's what that's how open router got started open router is growing rapidly and you guys are sort of having like a, you know, meteoric rise. So I don't know how the deal came together, but when you sat down with Patrick and John and thought through this, why did it make sense for you to like go ahead with the acquisition? And then why do you think it made sense for them? So I don't know very many companies other that are better than stripe at encouraging The growth of new businesses around the world, like really helping people create their own businesses and accept payments and get them off the ground. It's always just been a role model company for growing GDP in a bottoms-up way. And in a big way, we're trying to help do that for the AI community,

for the AI ecosystem, like helping people access models and helping people discover the right models for their use cases. This is one of the, I think, clearest mission alignments that we have with Stripe. We both believe that the best way of growing global GDP is by building heterogeneous economies that allow value to emerge from all corners of the world instead of just from one mega company, one single bank or one single model lab. And organize all of those heterogeneous sources into an extremely easy-to-use, safe, manageable, and useful developer experience and business experience. And that's a key thing that Stripe has been doing for payments for over the last decade.

And it's what we do for intelligence.

Alex AtallahSo a massive part of building businesses is not just going to be accepting payments. It's now sourcing intelligence.

It's the biggest line item, both for, like, new businesses' costs of goods sold and rapidly growing to be the biggest line item for any business on their, you know, operational expenses. So, you know, AI was the biggest contributor to GDP growth in the U.S. this year. And I think it's going to be, you know, a critical part of the economy in addition to accepting payments and how money flows between companies. Like, it will also be, like, how intelligence is sourced. And the two will become, you know, very tied. And that's a big reason it makes sense for us to do this on the same team with Stripe.

And, you know, like, we're, I think, fortunately, keeping the brand and keeping the vision and keeping the product and the roadmap and all the models and, of course, our model neutrality. That's something else both companies deeply believe in, is building really good, trusted, neutral infrastructure that, you know, across this extremely, you know, wild world, that we want to be heterogeneous, we want it to be diverse, and we want it to be, like, easy for new entrants to come in and innovate and get to market. it. You need neutral, reliable infrastructure to make that world happen and to avoid a world where there's just one company. There was a leaked investor letter that said they believe the singularity started on January 1st, 2026.

What do you think has fundamentally changed this year about sort of this like takeoff period we're if if you agree with that well without coming on the commenting on the letter i'm not asking you to comment on the letter that's uh i think something Because you see the data in like a really unique way. You're actually seeing like how people are using these models. What are they using it for? Which models? So if you go to our rankings page, there's a little button that says linear versus log on our tokens. And if you click on log to switch it to log mode, you'll see that it was growing until January 1st.

And then the slope dramatically picks up and has been, you know, consistently higher than it was up until January 7th or something like that. There really was a, like, change of growth, like, acceleration that happened around January 1st. And so in terms of timing. New Year's resolutions. People were like, use more AI. Yeah. Well, I remember, like, after New Year's or around New Year's, I was just like, oh, my God, like, I need to, like, generate some papers because I can actually brainstorm with the models now. And it's, like, kind of effective. Like, there's some, like, good ideas coming out of Opus.

And the coding was, like, so much better. I felt like we could refactor code and actually, like, count on it to work most of the time. And that was always my personal eval with models before, is like, hey, can I refactor this part of our code base in a way that makes sense structurally and have it still work? And suddenly that was working. So it was pretty incredible. And I think that it showed that we have to change the way we work. And so I told the whole company, hey, something insane has happened. We have to change the way we work.

Alex AtallahAnd if you're not using agents most of the time, if you're using autocomplete, it's time to switch.

No more autocomplete. We have to leverage cloud agents, but local agents are fine too. and um and we have to do a we have to better understand our velocity from a you know we have to like give ourselves more leverage here it's it's out there and the tooling is just primitive um and you know and so we started you know we like have launched agent tools we launched ori harness which helps people use any agent with open router really easily. Any agent harness. We started a lot of we started going very deep in tools and helping developers like get access to like server tools and inference agnostic or inference adjacent infrastructure that, like, does make your end outcome a lot better with better grounding or better relevance to your org.

And, like, you know, that... So in many, many ways, it changed the company. When the acquisition was announced on X, there was a lot of noise, a lot of comments, a lot of posts. And there was one take from people like Sam Lezen tweeted something like, there's no real technology behind OpenRouter. Every company is going to build their own. Ramp, which is a company focused on an entirely different set of things, launched their own router. What do you respond to that? Yeah, there's a lot of companies building MeToo routers right now. um and you know i i think like a big fundamental difference with

Alex AtallahIt's not a side quest for us. It is our main quest. Everything I'm 100% focused on is building the best marketplace, gateway, and router, and having that level of focus, I think, changes things dramatically.

um you know it means that we are ahead by you know like both in in the quality of the router the like breadth of the marketplace the um the like strategy that we set um and the talent that we acquire it's like it's the talent that's like focused on this problem um and you know and and we do a lot of work also with the ecosystem because like open router is kind of an ecosystem driven company. Like we want to help models and providers find their audience, like someone with a great innovation from like some Neo lab or some new inference provider, you know, we want them to find like to get to market as quickly as possible.

And we want the whole world to know about them. And we like our mission is to improve neurodiversity and AI. So that allows us to kind of provide natural marketplace benefits. The biggest marketplace has this natural moat where all the labs and providers and the supply side that has innovations wants to be on it. And in turn, developers and businesses find the best prices and the most variety and the highest quality inference all on open router and you know just making sure this marketplace magic works at scale yeah is like like the core focus of the business and it's you know like all marketplaces it's pretty hard to replicate and do you see a world where it really diversifies to like all type of models once there's amazing biology focused models and robotics models like people can go to open router to access everything yes and we have an enormous variety of non-language models um there's there's video models image models speech to text text to speech uh there's embeddings models there's re-ranking models and now there's like really exciting ways of comparing them all um we just launched this benchmarks explorer where you can explore models by benchmark and, and understand what the hell these benchmarks mean.

Like basically, like tons of people reference MMLU without having any idea what a single MMLU question looks like. Now you can see, and we also have, we created our own benchmarks too. Like we have image and video benchmarks where you can just see all models try to render a wine glass that's like half full. And you can see the models just, like, not quite get it. They have different levels of wine in each glass. But it's really easy to just see and compare them all. And we just kind of, like, find these cool, very general evals that help you choose what you want to choose and help your agent choose, too.

So, you know, you'll find, like, a very cool diversity of models that will get more diverse in the future. If I could ask you to put on like a pair of like future goggles and imagine this sort of future agentic economy where everyone has a ton of agents. We just ran into people on our way in here and everyone is talking about instinct and how now people can, they have their agents that can really make purchases and operate in the world. Plus like the payments infrastructure that Stripe offers. What do you think is this future? of economy that you or, you know, Patrick and John kind of imagine when you're piecing all these sort of amazing companies to work alongside each other, like even the acquisition of bridge that they made, there's sort of this very clear, like interesting set of things.

Yeah, I mean, I sometimes I talk to people about agent payments, and I'm and, and people are like, this is like, this is never gonna work. like I enjoy shopping. Like I want to go to Amazon. I want to like see the things that I'm buying before I buy them. Um, I would never use this. And, you know, I personally am not like that. Like I know I would buy a lot of stuff if agent payments worked really well. Um, I, I know I would finally get my toothpaste delivered on a schedule instead of like walking across the street to the pharmacy to buy it like every two months. Um, if I could do it like with an agent and like cancel the schedule or modify the schedule with just text um and i know i would even do that for like shirts and we were just all kinds of things we're the same shirt that you were six years ago on our podcast you could get a few more shirts i don't have like that many clothes i would definitely get more clothes if my agent like knew my size like you know because i ordered something within the past and like it knew some things about me and knew some things about brands that like matter around the world. And there are studies that prove that people enjoy the process of planning a trip more than the actual trip. Like some people find a lot

of gratitude in those experiences. I am not one of those people. I, there's still some experiences where I might want to like do it manually, but with a lot of things in AI, you know, there, there's the span of tasks that you know you do and love and then there's the span of tasks that you don't love and you don't even know how wide that span is because you don't do those tasks like you don't understand the magnitude of how big that section is because you avoid it naturally So if an agent is going to take over that whole set of tasks for you, it, I think, is going to be way larger than everybody thinks because everybody's underestimating it.

Alex AtallahYou don't know the magnitude of purchases you're going to make for things you don't like to purchase because you don't do it.

So you can't know how big it is. Do you think people will buy more? Yeah, yeah. Yeah, I think people will buy a lot more and people will buy things they don't normally buy and people will programmatically buy things that they don't normally buy or create, you know, like group buying entities that like buy everything for a whole group. group, people will give more gifts, people will use more of their disposable income, people will donate more, people will possibly return more goods if it gets easier. And so there's all kinds of magnitude of the 1% issues, where once you build a really good product that targets the 1% of use cases that people aren't doing, that 1% blows up into a much bigger percent.

This is constantly happening with AI, where the top 1% of intelligence tasks that people aren't able to do themselves suddenly blows up to become like most of their spend because it's now just so much more efficient to like do R&D and discover. and spend there. This past week, OX Alpha went a bit viral, which was basically an anonymous model. No one knew who made it. Now we figured out that it's from Z.ai. What did, watching that experience, what did that prove to you about like model preferences and how much people care about brand and some of these things? yeah so when it was in stealth mode which is a program that we have that we started back in 2024 and why did you start that so in in uh what uh what we we worked with open ai to to do the stealth model program for the upcoming launch of gpt 4.1 and we wanted to provide, we wanted to basically let users evaluate a model without showing bias around the brand that made it and get a better, a way of providing interesting feedback to the model lab, but also to create a way for people to build their own benchmarks and see how the stealth model performs on community benchmarks, benchmarks other than the things the model labs normally evaluate

internally. So it was a good way of kind of doing revealed preferences without implicit bias that people have. And we've continued the program and done it with many model labs since then. And like this past month, we did it with Z.ai for their launch of GLM 5.3 Flash. The model was called Ox Alpha. and or yeah ox alpha and a lot of people tried it more unique users tried it than any frontier

Alex Atallahmodel um it was very very popular it was the largest model launch in open routers history

in terms of tokens and it was just very capable um we also offer the stealth models for free and a lot of developers just found it like very capable at long horizon work and very good pros as well. And when it exited stealth mode, it's still pretty cheap compared to other models, right? Yeah. GLM 5.3 flash is very, very good cost per task performance. And I believe they did about five trillion tokens in their first four days, which is very, very high after the model was revealed. A lot of the industry is very focused on cost per token, like price per token. We're guiding the industry towards thinking about cost per task and cost per session because the models have such different profiles around token efficiency and it, you know, it matters like how much it costs to complete the task, not price per token.

So if you look on our rankings page or on our model pages, you can now see like estimates of how much it actually costs to have like a reasonable session in cloud code or codex or any agent using each model. and if you go to that spot in our rankings page today you can see glm 5.3 flash like ranked really high wow um it was also just cool to see that like cultural conversation around it everyone was so excited and curious um and that was very interesting um if you were to look very far out do you think people will just like pay for an outcome like they like this will all be sort of like electricity you don't care where it's made or how it's made but you just are like oh i'm willing to spend this much on getting This outcome, and then you guys handle all of that, and so people don't care about brand, or frontier model, or Chinese open weight model, or these different ideas that are a big deal right now.

I think there will be markets for both people who care about the model and kind of intelligence that is being used behind the scenes, and people who don't. A lot of consumer apps, for example, people don't really care about which model they're interacting with under the hood. They just care about the job getting done. B2B is a little different. Businesses do tend to care about the model and actually, like, want to understand and build routers themselves. Like, I believe most companies are going to become model routers of a new type. And, you know, we'll see, we've been saying this since, like, 2024.

Like, a lot, like, all startups, all software startups will become model routers. They want to know which intelligence inputs they're using, and they want to understand how those choices affect their cost of goods sold, their COGS, and their key performance indicators, their KPIs that they're trying to maximize for the product. And choosing the models and understanding how the models are prompted are the inputs to those equations. And so you want control as a business, but you don't really care as much as an end user sometimes. And do you think that, like, I think there's a debate right now whether it's bad if we widely use Chinese open-weight models in the U.S.

Yeah. The question is, why would it actually be bad if you had to, like, straw man the other side?

Alex AtallahThe safe way of using these models involves companies creating evals, which are like tests that they run on their tasks that they care about ahead of time, and making sure that the models are performing well on these evals.

And this is like the critical act that businesses will have to do in the next 12 to 18 months. In these evals, you will find exactly what that means for every model, not just the Chinese models. and um i think that's like a foundational case that i want to make um and it doesn't have to do with whether the model is chinese or not now if there's some kind of like geopolitical game being played by china where like all the model labs become like slightly more likely to do something or all the the chinese models become you know slightly more dangerous in some dimension um we do want to find that and like i you know i think i think the we'll have a lot of companies that go look for those outcomes um and just make sure that everybody's secure and some of those outcomes might be completely unintentional it might be like you know somebody trying to feed something to the model labs training data pipelines to get models to do something there's a lot of model epistemics to worry about, where people game the AEO, the agent optimization that allows models to have some kind of preference for some company or some brand or some ideology.

Companies and nonprofits will form to help look for that stuff and it will become a huge deal in the next you know two years I think it's it's gonna be like a big problem and very very interesting when people find like key examples of this happening and you know I'm really excited about like new innovations or new projects that are working in that direction that's okay so they'll just be a new set of companies and tools and things built around around this? Yeah, kind of like how, you know, there was like a McAfee antivirus, like a industry that, that emerged around keeping people safe from random things being downloaded on the internet.

You know, you don't want to ban the whole internet just because there's some bad things that some people can download somewhere on the internet. You want to create an industry that protects people at the layer they're operating in or build operating systems like Apple has that provide really good safeguards at the user level. And how do you think the web changes over the next 10 years? Is it going to be optimized for humans or for agents? Well, we're seeing pretty massive usage of the web by agents in terms of bytes transferred And, you know, like, I think a lot of companies are going to realize that, like, optimizing for agents becomes their fastest growing way of growing their payment volume.

The more, like, experiences that exist where people can buy things directly from an agent, the more it makes sense to optimize for agents. And the more people optimize for agents, the more companies will start up. They'll let people buy things with an agent. And so those two components are like a virtuous cycle upwards of more and more agent GDP. That's the driver that I think will cause agents to become potentially the biggest architect of GDP growth. and i i think like humans there will still be like a big market for like the human economy but you know it would make a lot of sense i would not be at all surprised if agents you know are tied to most gdp growth in the near future do you think this is sort of the the situation where like crypto makes the most sense like it will find like people will be using stable coins more than just you know across country payments and sort of where it's being used today well i think the the stable coins that exist are probably better for agentic payments in part because they are so much easier and so much more deterministic to manage than dollars, and I think we'll have like a lot of emerging use cases where the ease of converting to a stable coin becomes so, so high that consumers don't even know that stable coins

are powering the product, and they can use their funds wherever they are, but if they use stable coins directly, they get potentially better economics and lower fees, and businesses use that to kind of like go down and incentivize people to move to stable coins as well. So I think we'll see more adoption. What do you expect for GDP growth? Like what we've seen the past couple of years versus the next three years? Because there's this idea of like, have we already achieved AGI? Is it happening soon? Is it happening in a year? There's a lot of discussion about recursive self-improvement and like these sort of potential mass.

the steps and then how does that translate to the way we can use agents and how much we trust in them? I think these model advancements will make agents more trustworthy, more effective and more generally capable basically so more trustworthy means that you know for the tasks that I know they can achieve I feel like wow this is just going to be so reliable like I know it's done this many times before my friend tells me it's done it many times before I know it's going to do it right away I know it's going to do it a low cost I'm going to give it to an agent. More effective means that when the agent does it, I end up getting a lot more out of it than I was in the previous version of the agent or the previous pre-RSI models.

And then more generally intelligent, I can just give it more tasks. And my agents can give it tasks too. And it can be used in pipelines. It can be used as a component to another business that uses a component to another business and you can stack these components much more reliably and drive the price down of each layer in the value chain. What do you think are the last things that we will entrust agents with? What I mean is what tasks are humans still key to and so much better at? In your job, are there things you've been able to offload over the last two years? And when I mean offload, I just mean hand to an agent to actually carry it versus like what does Alex have to be at the core of?

There's probably going to be a lot of value in the things that humans like to do with each other. And that humans like to own themselves and will pay to own themselves. You know, they actually want to be the approver rather than having an agent be the approver. They want to own the brand. They want to say they created the brand. They want to say they, like, you know, designed or envisioned some concept. They want to, you know, they want to claim the fame or claim the credit. And, like, I think, you know, a lot, it'll be, like, humans, I think, will probably gravitate to those things because agents are taking on so much of the actual knowledge work and and and software work and eventually hardware work with robotics um and uh and and and like it's not all dystopian like humans will basically move up the stack of value creation i think continuously where they're just always trying to think of like well what are the agents not doing like what are the models like not thinking of i mean they're not thinking of like the things i like to do um the things that are going on in my brain and uh how can i like help people who have similar experiences to me so i read an old blog that you published and in 2024 you wrote that um engineers will become

scientists and that humans keep an edge by acquiring context from non-digital places uh do you still agree with that and where do you acquire your context yeah i do still agree with that i think uh a lot of our engineers for example have been writing papers basically in our blog about benchmarking uh models and web search providers and and like Different open-weight model providers along different, like, inference quality dimensions, and, like, being a rigorous scientific thinker is now, like, very in demand. Like, it's a very important part of OpenRouter and very, like, big pillar of the company is extremely high-quality benchmarking that developers can build on top of.

So they can count on OpenRouter to benchmark and find the highest quality inference and the highest quality, you know, grounding, the highest quality models for their use cases. And then, you know, we encourage developers to build their own benchmarks and evals on top. And I think that's, like, how value stacks up, where, like, you don't have to start on floor one of the building every time you can start on floor 10 and build upwards more quickly. So I think like being a good scientific thinker, like knowing how to, you know, steel man the other side of, you know, an argument that you're trying to make, knowing like how, like understanding variance and, you know, confidence levels and, you know, potential weaknesses in, you know, in the, you know, the end conclusion.

like these are all things that are muscles that you need to like train yourself to get better at and um and like we have engineers now like doing that work um and we're also hiring you know traditional researchers as well um I think other companies like us will probably end up doing the same thing you co-founded OpenSea and you're building OpenRouter today I think you're someone that is very good at like diving deep and really seeing these like trends of how develop I mean open to is more consumer but like at least how developers you know work and what was missing um if you were to look forward and this might be an idea for you know someone else listening that wants to start a company are there any other things where you're like the next open is what and and you know in what sector well I think the next three to four years are gonna be really insane a lot of things are gonna change and and it's it's It's really hard to make predictions right now because we're still seeing significant intelligent advancements every four months.

Alex AtallahOne way of looking for companies is to figure out what the bottleneck is between the best

possible answer and where you think answers from models are going to wind up in 12 months not just answers but actions and services and like any kind of you know any kind of like monetizable thing that a human might pay for or an agent might pay for so you know there's still a lot to be done in robotics like Like, we don't have a consumer robot that everybody has. Seems crazy. Like, why is that? Like, there's a lot of questions you can ask there, and there's a lot of ways to go deep. There's also some promising companies, too. There's, like, context. Like, how can we, like, get... No. Models better and more continuous context to help answer their questions.

When models get better at memory, what will be needed for a model that has really, really good, even perfect memory about everything that has happened in the past? What kind of infrastructure will be needed? What kind of connections will be needed? What kind of collaboration will be needed between humans and agents? um you know i think you can also um you can also spend time thinking about like a lot of tasks will be very you know solved and deterministic and we'll have you know models that are like good at doing these specific things but many tasks will be like non-deterministic and all researchy and all very like you know yeah like figure out some incredibly valuable incredibly difficult problem that like would have taken you know 20 000 20 million dollars in inference to do um but let's try to do it in like a day like thinking about those problems and thinking about um what happens when it all becomes doable within a day or within you know a week um is really hard like i like a lot of people don't know they can just ask chat gpt or, you know, most agents to read all the news sources they care about and send them a summary every two days.

Like, you just have to ask the question sometimes. Yeah. So I think, like, being very curious is just a good general tactic and also thinking about, like, the little portion of tasks that get unlocked at the end of the value curve is another useful tactic. I interviewed a professor who has actually worked with some of the labs on analyzing how people are actually using their models. And he had a takeaway where it's like so much of the model improvement isn't about the model improvement. It's about that user really learning the model and how they prompt it. And I still think there's a big gap there between the way most people know how to interact with a model and how to prompt it and how to use it best.

It's not actually super obvious to most people. It's not perfectly created for humans. It's like you have to learn it. And it also is interesting as the models improve and change so quickly. Do you have to keep redoing that process? And as people get super attached to certain models, I do think it's going to be interesting. like if they are um improving really fast do they feel a change when it's like you know since they're nine years old they they use this model to talk about their problems like how does that model maintain a lot of consistency in the way it interacts with that person and these ideas are interesting yeah um yeah people have like like very personal conversations with lots of these models and like it is what do you think about that extremely cool to see somebody else's chat gpt session when i'm like with a friend and just asking them like what what is your chat your your like clod or chat look like and it's like peering into somebody's mind it's crazy by the way it's it's really intense i had a friend that um their boyfriend read their chat gpt history like instead of reading their phone and i hope they don't watch this but actually that is so much more invasive than any text you could ever read because most people are not

yeah you know it's it's appearing into someone's like actually deepest Questions and qualms and truth-seeking about a relationship in kind of an insane way. And I have friends at the labs that are shocked with the level of intimacy in which, like, I open up to my Chachi Biti and I need to not. But I just find it to be this, like, it's like, I don't know, it's this free source of someone being able to take in a lot of information and spit it out in some type of, structure, which I think feels really good for people. And so often when you're talking to someone, that's actually what you're looking for, like being heard or for someone to make your problems feel like structured, doable, and have something to move forward with.

So it'll be interesting. And on this point, because like OpenRouter, not everyone is using it for coding, like people are using it for other activities as well. um there's going to be you know a set of lawsuits i'm sure around even what happened with character ai like when people become deeply attached to models and you know self-harm or do dangerous things and people go find that they've had conversations with their ai about these things how do you think about like model accountability in in these situations because i'm sure this will be an upcoming um you know an upcoming issue obviously you know the stakes are really high for some types of conversations to like get it right for the model like i think people you know when you have a lot of self-doubt um when you don't trust your instinct instinct, you either spiral and go nowhere or you ask for help, you ask a human for advice.

And in addition to providing lots of good structure, the models also provide a way easier way of relieving your self-doubt, feeling like you fact checked and gut checked and and vibe-checked your opinions and your instinct. And, you know, it replaces a lot of human-to-humor and interaction and human-to-human advice. Humans make tons of mistakes, too, when they give advice to people. Of course. It's so based off of the thing that just happened, too. Like, we can't help but put our own biases into our advice for others. Yeah. I mean, humans make tons of mistakes, too. probably on average like the models will make fewer mistakes than humans um but what they don't have that humans do have is accountability like a human you know is on the hook for like the advice they give someone who's really struggling and um you know it like can take responsibility and can like you know go in person to meet with them and there's a lot of like in-person interaction the models can't do too so and sometimes I found like I'll prompt it to push me so I'll say do you I want it to push back on me because before I haven't been using it as much recently for these things but you know throughout different breakups I definitely texted me a lot and it was like you

know push back on this this this way I'm looking at is there something I'm missing and then it will change its response I'm like well did I have to tell you to you know what I mean before it was pretty affirmative yeah and i mean obviously i'm biased but i think one of the coolest things to do is to see how different models respond to you differently and to get like a panel of experts what are the vibes of the different models if you had to i'm sure you haven't gone and like i don't even know if you ever open up to the models i mean totally outside of coding putting coding and building things aside um i found fable to be the best at brainstorming personally or it just feels like it it's drawing on ideas a little it's really hard to benchmark So like it's hard to know if I if I just don't have some like a little like bias towards it.

It is the most expensive model that makes some people think that it's the biggest one as well. Like bigger models I would think would have like an advantage of being a little more creative and pulling like you know crazier ideas out of thin air. And so it would make sense if I have a bias towards fable at just being better at drawing creative ideas gpt 5.6 soul is much easier to control the language it feels a little bit more steerable um and um kimmy k3 i think also has like very good tone and voice and is similarly steerable um you know one time my mom was like trying was trying to figure out what was wrong with my dog like none of the doctors knew she had a neurological issue and wasn't able to walk and i just put all the symptoms into those three models and like sent them all to my mom and i was like hey you know which of these sounds best and you know i was surprised that she picked kimmy KimmyK3. Or it was KimmyK2.5, actually. But it was, you know, it shows that, like, the voice matters a lot, even when, like, all the ideas are, like, pretty uncertain.

And my last question is this. In our life, like, especially outside of San Francisco, when I'm in LA or even New York, I hang out with my friends, and their feeling genuinely is, like, i hate ai because i didn't ask for this dystopian shift to take part in in the world like i like the world as it is i really like my life and everything that most people hear about it is like job loss ubi like you're in your dystopian like ai media loop and that's all you see um what do you actually like believe the future is gonna look like like what excites you yeah my optimistic take is that we underestimate our appreciation for like human creativity when we can identify it for example when the dune 3 trailer came out it was sent to me in a slack message so i watched the trailer from slack and it was in a context where it could have been a fan-made trailer i wouldn't have known like i didn't know who i didn't know the trailer was like from um the the studio and i i like clicked the button and i was like okay this is probably a fake you know fan-made ai trailer and then it got to the soundtrack like this the sound that i've just never i've never heard anything like the sound before it was like a crazy new dark vibe

that was very different from the first two Dune movies. And I was like, oh, my God. There's no way this is a fake trailer anymore. Like, this is, like, very creative. This is, like, I've never heard anything like this before. That was an incredible moment because I was like, oh, my God. You know, I can tell that this part is, like, some human. Maybe they used, I doubt they used AI, but, like, in the future, Maybe they'll use AI to create components of it, but the overall architecture, if we can see, oh my god, I've never heard or seen or felt this way before, that novelty chasing is going to be a bananas race and a really big thing and big deal for creative industries like Hollywood.

That's really interesting. I've actually never heard that take, but I think that's really cool. Thank you. It's very hard for AI to verify and like get really good at creativity. I, you know, I don't know how, like, I think a lot of researchers believe that creativity is something you can get to with reinforcement learning, just like, you know, executing tasks better. Maybe, maybe we, maybe we get there. I don't know. But yeah, models fold into the most obvious, like, assurable, like, assuring response. That is sort of in competition with the idea that the most creative pieces of art came from something weird and unique and, like, non-obvious.

Yeah, yeah, yeah. We'll see. Humans are still the weird freaks for now, so. Anyway, Alex, thank you so much for coming on the show. Thanks, Sabrina. And thank you for taking the time. Thanks for inviting me. It was great to be back. If you enjoyed this episode, please take a minute to subscribe. It really helps me grow the show and continue to put out new episodes every week.