Real Alpha
Watch modeSubscribed

MTS · 3 min · Oct 17, 2026

AI solved research problems Aaronson had left untouched for 25 years

SA

Scott Aaronson

Theoretical Computer Scientist, UT Austin

Quantum computing theorist who spent two years on AI safety at OpenAI and writes the Shtetl-Optimized blog.

  • Theo JaffeeHost · Co-founder & host, MTS

    One of the first members of a16z's new media team; earlier hosted the Emergent Ventures-backed Theo Jaffee Podcast.

  • Sophia DewHost · Host, MTS

About this segment

Aaronson says AI solved several old research problems in one shot, work he estimates might have taken him at least 100 hours. He also reflects on how relying on AI can accelerate the atrophy of his own skills.

From “OpenAI Just Solved 370 Major Math Problems | Scott Aaronson”, Oct 17, 2026. Watch the full episode

“some of the open problems for my papers 25 years ago, and just boom, boom, boom”
SAScott Aaronsonat 34:48 · MTS34:48

Takeaways

  1. He gave AI open problems from papers he wrote 25 years ago.
  2. The problems were ones he might have solved with enough time.
  3. He uses AI for research help, writing, graphics, and everyday questions.
Watch 3:29 from 33:57

In this episode

Full episode
  1. 14:12AI crossed long-standing limits in multiplication and Fourier transforms5:32
  2. 4:26Mathematicians need to understand AI proofs, not dismiss them8:13

  3. 33:57AI solved research problems Aaronson had left untouched for 25 years3:29

  4. 3 more moments in the full episode
Up next · FoundersMeta bet on personal agents while rivals crowded into coding3:42

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 34:03

    using AI for math and for other work, you know, administrative personal life or whatever? yeah no i mean i use it i uh uh you know i i i'm i'm not super sophisticated i mean in how i use it like i know people who have you know a whole team of agents you know just doing tasks for them you know overnight you know they wake up in the morning and uh it's like they have this that you know this whole swarm that's uh built something for them you know i mostly just use astra you You know, I just use the, you know, basically my GBT Pro subscription, and I ask good questions when a question occurs to me where, you know, its answer would be useful.

  2. 34:47

    So, you know, I did feed in some of my, you know, this summer, I fed in, you know, some of the open problems for my papers 25 years ago, and just boom, boom, boom, you know, it solved them in a one-shot way. i'm planning to do a blog post about that you know soon but uh um yeah i mean you know like these are problems that probably i could have solved if i had spent enough time on it you know but i didn't right uh what is enough time like 10 hours like 100 hours yeah yeah more more like 100 at the least you know and they would have to be like you know i mean the problem is that these days like i don't really get contiguous hours to think about reason right i'm just constantly interrupted by you know needing to answer emails needing to deal with my kids needing to uh um you know deal with commenters on my blog needing to deal with you know just you know administrative

  3. 35:49

    stuff with with travel right and so so you know you could say already before ai like i was moving to a moment where like less and less of my time is just thinking hard about research More of it is, you know, mentoring my students, you know, sort of, you know, helping to guide them to solve technical problems, right? Yeah. And so for me, AI just kind of accelerates that in a certain sense, right? Like it accelerates the atrophying of my own skills.

  4. 36:19

    uh and and you know when i have like you know a a researching type of problems like you know it is it is so tempting to just you know you know if it's something that you know has an answer that you could work out in some hours but you know gpt can just work it out for you right now right that is that is very tempting to do and i often do it uh so i do use it in that way you know i've also used it for transcribing you know i've used it for uh for help with with with la tech you know with type setting i've used it for making graphics for my for my talks you know so like all the you know i've used it for medical advice i've used it for you know various kinds of life advice for uh you know uh for just like general curiosity about history and science so like you know the whole range of things that you could imagine using

  5. 37:15

    it for. Yeah. I do think... Oh, do we have to wrap now? Yeah, we have to wrap. But it was really great