Daksh Aggarwal

Math PhD student @BrownUniversity

[Nate Gillman @ CVPR @GillmanLab]

Ever wish you could tell a video model what to achieve, rather than just how to move?

Introducing our CVPR 2026 paper, Goal Force!

Instead of simulating a direct push, our model plans the entire causal chain (the "how") to achieve your specified goal (the "what"). 🧵(1/n)

[Chen Sun @jesu9]

Can you inject causal control to models trained solely on video data? How about eliciting their understanding of physical forces when animating a photo? Nate's latest project shows that both are promising directions to explore! I tried it on my favorite photo, and you should too! https://t.co/UjFv3LQVYC

[Nate Gillman @ CVPR @GillmanLab]

Ever wish you could turn your video generator into a controllable physics simulator?

We're thrilled to introduce Force Prompting! Animate any image with physical forces and get fine-grained control, without needing any physics simulator or 3D assets at inference. 🧵(1/n)

[François Chollet @fchollet]

People who treat prediction markets as some kind of magical oracle are hopelessly lost. Even the valuation of very high-liquidity public stocks is anything but efficient. Now, election bets? Forget about it.

[François Chollet @fchollet]

Remember that prediction markets are very low-liquidity -- with a couple M$ you can clear the entire order book. They do *not* encode any kind of "efficient" prediction of future outcomes given all available data. They encode the uninformed opinion of a handful of rich gamblers.

[Chen Sun @jesu9]

.@GillmanLab and the team have done it again! 🎉 Fourier Head brings theory-driven enhancements to LLMs when modeling continuous tokens! The best part? It actually works and is easy to implement!

[Nate Gillman @ CVPR @GillmanLab]

Check out our project page for more details (paper, code, visuals…)

https://t.co/V4fHQfqDxU

This was a team effort :) joint work with @dakshces @mik3fr33man @saurasingh @jesu9, a fun collaboration between @BrownUniversity / @GoogleDeepMind, and of course, Fourier's head (n/n)

[Nate Gillman @ CVPR @GillmanLab]

LLMs are powerful sequence modeling tools! They not only can generate language, but also actions for playing video games, or numerical values for forecasting time series. Can we help LLMs better model these continuous "tokens"? Our answer: Fourier series! Let me explain… 🧵(1/n)

[Nate Gillman @ CVPR @GillmanLab]

Check out our @ICMLconf poster tomorrow if you want to learn how you can improve human motion generation performance by injecting expert physics knowledge during training :) presenting w/ @jesu9 + @dakshces

I'll be in Vienna through Sunday!! #ICML2024

https://t.co/VwvwDBKtdv

[Chen Sun @jesu9]

Now accepted by #ICML2024, code is available at https://t.co/L38OsDzZrB !

[Chen Sun @jesu9]

Video genmos (e.g. #Sora) are touted as "world simulators". Can we leverage their outputs to create more powerful genmos? Our paper https://t.co/27pRNhPA82 demonstrates mathematically that a naive solution would lead to model collapse, but a self-correction op would fix that!

[Nate Gillman @ CVPR @GillmanLab]

Excited to share our latest preprint: “Self-Correcting Self-Consuming Loops for Generative Model Training”. It's a step towards generative AI models that can learn from the universe of data they generate!! 🤖(1/n)