michael tetelman

PhD, AI Researcher at BayzAI. Bayesian Methods for AI, AI Fairness, RL for AD+.

[michael tetelman @mtetelman]

@zdeborova @zdeborova , you made a non-trivial statement. Non-memorization means with all other training conditions equal it learns a best approximation of a true distribution, which gives a best generalization. So, you are saying it is not? Do you have an example?

[michael tetelman @mtetelman]

@zdeborova Could you point at studies that analyzes a difference of generalization vs non-memorization?

[michael tetelman @mtetelman]

@andrewgwils DL theory that explains could be useful but what we really need is something more practical, that excludes guessing from finding new optimal architectures.

[michael tetelman @mtetelman]

@hr_unhinged Good job Karen. Do not forget to charge yourself overnight.

[michael tetelman @mtetelman]

Priors are all you need!

[Sanjeev Arora @prfsanjeevarora]

Insightful blog post from @ShunyuYao12 about what's needed for next level of AI progress. He's been driving AI progress since his work @PrincetonPLI including SWE-Bench.

Related but somewhat divergent view from @karpathy who argues (correctly in my opinion) that we need to go beyond current simplistic ideas in RL. https://t.co/qqvpVCSGh2

[michael tetelman @mtetelman]

@andrewgwils Nah. Until you have a chip in your brain and become a network node you will have to do all that yourself.

[michael tetelman @mtetelman]

@2prime_PKU Could Adam tell how these architectures were optimized vs parameter optimization for fixed architectures?

[michael tetelman @mtetelman]

@karpathy Bootstrapping RL with a world model is a way to successful scaling.

[michael tetelman @mtetelman]

Cool!

[Fenil Doshi @fenildoshi009]

🧵 What if two images have the same local parts but represent different global shapes purely through part arrangement? Humans can spot the difference instantly! The question is can vision models do the same?

1/15

[michael tetelman @mtetelman]

@andrewgwils If adjusted for inflation it's actually an axiom.

[michael tetelman @mtetelman]

@WKCosmo Ha-ha, here is another real "elegant" formula:
abs(e^pi - 20 - pi)<10^-3

[michael tetelman @mtetelman]

Transition Matching (TM), a novel discrete-time, continuous-state generative paradigm that unifies and advances both diffusion/flow models and continuous AR generation:

[Tanishq Mathew Abraham, Ph.D. @iScienceLuvr]

Transition Matching: Scalable and Flexible Generative Modeling

"This paper introduces Transition Matching (TM), a novel discrete-time, continuous-state generative paradigm that unifies and advances both diffusion/flow models and continuous AR generation. TM decomposes complex generation tasks into simpler Markov transitions, allowing for expressive non-deterministic probability transition kernels and arbitrary non-continuous supervision processes, thereby unlocking new flexible design avenues."

[michael tetelman @mtetelman]

Integration Flows: a novel generative modeling framework

[Tanishq Mathew Abraham, Ph.D. @iScienceLuvr]

Integration Flow Models

"Integration Flow is the first model with a unified structure to estimate ODE-based generative models and the first to show the exact straightness of 1-Rectified Flow without reflow."

[Yoshua Bengio @Yoshua_Bengio]

The most advanced AI systems will likely exist first within the companies that developed them, with little opportunity to collectively prepare for their impacts. This dynamic creates unique risks highlighted in this recent paper by @apolloaievals and should become a growing priority for AI policy.

[michael tetelman @mtetelman]

Can we cook neural nets?

[michael tetelman @mtetelman]

@andrewgwils Simple models of the world and wish for quick rewards is a road to disaster. That's why principles are more important than models. Principles are wisdom derived from ages of human history. We have to stick to it to make good and make it right.

[michael tetelman @mtetelman]

@andrewgwils That may change your future in unpredictable way.

[michael tetelman @mtetelman]

@elonmusk Yep. Leave your entropy outside, then enter.

[michael tetelman @mtetelman]

@docmilanfar That's so much fun: having a bottle that's never full. But what it means for images? Are there any integral topological invariants on image space?

[michael tetelman @mtetelman]

@docmilanfar Perfect benchmark for robotic VLM