The alpha, twice a week
That was 2 min.
Get the best new moments in your inbox every Wednesday and Saturday. It's free.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
MTS · 2 min · Oct 20, 2026
Clear verification helps AI advance faster in math and coding
Nenad Tomašev
Senior Staff Research Scientist, Google DeepMind
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
Math and coding offer proofs, tests and error traces that give models clear feedback. Subjective fields lack that loop, making comparable progress harder.
From “AI Science Requires Physical Labs To Scale | Nenad Tomašev”, Oct 20, 2026. Watch the full episode
Thanks. We'll look at and fix what's off.
“So in mathematics, there are formal specifications and formal provers that can check and give definitive feedback to the model, whether it's done something right or not.”
Takeaways
- Formal provers can check whether mathematical work is correct.
- Coding tests give models feedback to improve their outputs.
- Qualitative fields lack equally definitive feedback.
In this episode
Full episode- 2:04AI discoveries need physical labs to become scientific progress2:42
- 11:57AI alignment should leave room for social values to evolve3:39
- 21:43Agent swarms need structure to stay manageable3:54
- 5:13Cognitive friction can help people keep their skills sharp4:09
9:57Clear verification helps AI advance faster in math and coding2:03
- 15:33Personalization can make models agree when they should push back2:21Up next
Transcript
Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.
- 10:00
unprecedented capability gaps in what AI can do, and I'm curious what you think the cause of that is. Is it that we haven't sort of assembled the right mixture of experts and brought them in-house into Frontier Labs? Is it kind of the opposite where all of the Frontier Labs or like have the talent monopoly and so there's like other nicher fields that end up not getting to capitalize on that kind of like really ai verse talent what do you think is is the problem here um i mean we're talking about uh the entirety of um like the full set of human capabilities and everything that we have in society like at the end of the day so that's simply a lot of stuff And in terms of where the capability gaps are, I think maybe uncontroversially, even though I cannot speak to all of it since I only work in one section of that spectrum,
- 10:56
where we have seen the most advances recently, mathematics being an example of this with many big results that were proven, are domains where we can have formal and computational verification. So in mathematics, there are formal specifications and formal provers that can check and give definitive feedback to the model, whether it's done something right or not. Similarly, in coding, I mentioned the genetic coding, there are unit tests, integration tests, all sorts of tests that can give feedback or error traces and help improve systems.
- 11:31
Whereas in some other domains of interest, I mean, for us as humans, there is no such definitive feedback the judgment is more qualitative more subjective maybe people will disagree so in those kinds of domains it's again maybe somewhat uncontroversially harder to make leaps in the same fashion just because yeah it requires more effort and time to do so yeah totally so