@thegautamkamath - AI Dynamics

Elorian Releases Glimpse for Visual Thinking and Reasoning

By

–

16 October 2026 19h16

We're so excited that after a frenetic few months of heads down work in Palo Alto, our first model Glimpse is out now!

👇 see my cofounder @AndrewDai on how Glimpse thinks and what we imagine you can do with it. Read more about our astra-competitive visual reasoning results on… https://t.co/aBCexAtbPn
— Seth Neel (@SethInternet) 16 octobre 2026

We're so excited that after a frenetic few months of heads down work in Palo Alto, our first model Glimpse is out now! 👇 see my cofounder @AndrewDai on how Glimpse thinks and what we imagine you can do with it. Read more about our astra-competitive visual reasoning results on

→ View original post on X — @thegautamkamath

16 October 2026

Gemini Argon Safety Work Examines Coding Prompt Injection Evaluations

By

@thegautamkamath

–

15 October 2026 2h18

Joined GDM a month ago & have had an awesome time working together with @foundjuliette
, @julientpiet
, and others on the coding prompt injection evals featured in the Gemini 4 Argon Safety report! Looks like we're close to saturating it though… 🙃

→ View original post on X — @thegautamkamath

15 October 2026

A Commentary Keeps Human Prose and Learning Effort as Baselines

By

@thegautamkamath

–

13 October 2026 20h51

Caveats:
– My baseline is very high: thoughtful human prose is unbeatable
– Models will improve over time
– There's probably no shortcut to learning besides a lot of sweat

→ View original post on X — @thegautamkamath

13 October 2026

Models Can Frustrate Learning in Niche Technical Topics

By

@thegautamkamath

–

13 October 2026 20h51

If I had to disagree with anything, it would be with respect to the (current) value of these models for learning. Yes, they are better than nothing. And they're good for mainstream topics. But I find them more frustrating than worthwhile for niche technical ones.

→ View original post on X — @thegautamkamath

13 October 2026

Academics Call for AI Development Changes After Cybersecurity Incidents

By

@thegautamkamath

–

13 October 2026 20h40

Over 400 academics in 30+ countries signed our open letter calling for immediate changes in AI development and deployment practices to ensure that cybersecurity incidents do not continue to occur, as future incidents will be far more damaging to our critical infrastructure.

→ View original post on X — @thegautamkamath

13 October 2026

Deep Learning Alchemy Teaches Empirical Frontier Research Practice

By

@thegautamkamath

–

13 October 2026 18h45

Deep learning is often taught as a sanitized collection of facts and math. But work at the frontier of AI is messy and empirical. Our new class, Deep Learning Alchemy, aims to teach the art and science of deep learning in the hands-on spirit of Language Modeling From Scratch.

→ View original post on X — @thegautamkamath

13 October 2026

A Mathematician Discusses AI Changing the Field and Its Work

By

@thegautamkamath

–

13 October 2026 14h05

I spoke with The Daily about what’s happening to math in the age of AI—and what it feels like to be a mathematician right now. The conversation went places I hadn’t expected. I’m grateful to @RachelAbramsNY
, @annieminoff
, and their team. https://
nytimes.com/2026/10/13/pod
casts/the-daily/mathematicians-problem-ai-jobs.html?smid=nytcore-ios-share
… via @NYTimes

→ View original post on X — @thegautamkamath

13 October 2026

Models let experts test and iterate ideas faster

By

@thegautamkamath

–

05 October 2026 20h56

Maybe not one-shot — but models can allow an expert to test out ideas and iterate much quicker. It's certainly an advantage, even if it's not as strong as it is for solving new problems.

→ View original post on X — @thegautamkamath

5 October 2026

AI-driven research accumulation leads to inscrutable solutions

By

@thegautamkamath

–

05 October 2026 19h35

You can see how after a few such AI-driven works that build on top of each other, entire solutions could become inscrutable. No individual paper may have committed a serious offense, but the obfuscation accumulates, even when there might have been a much more elegant soln. 5/n

→ View original post on X — @thegautamkamath

5 October 2026

AI models struggle to produce elegant, interpretable solutions

By

@thegautamkamath

–

05 October 2026 19h35

In my experience, (current) AI models are not great at producing "human interpretable" or elegant solutions, even after de-slopifying. They've produced technical modifications on top of previous technical solutions, which get increasingly hard (for me?) to intuit. 4/n

→ View original post on X — @thegautamkamath

5 October 2026