Sushrut Thorat

🚨 SOTA shape-biased ANNs, that experienced their datasets through the developing eyes of humans (from birth to 25 year olds). There's still hope for cognitive (neuro)science knowledge equipping ANNs with cool new abilities which seem elusive from the pov of ML research.

We invite postdocs interested in faculty positions to join us for an interactive Young Researchers Meeting on 2 August 2025. Hear from the Director, Deans & faculty members, and find out what IISc has to offer! Register by 12 noon (IST) on 1 August at: docs.google.com/forms/d/e/1FAI…

Terence Tao on the supposed Gold from OpenAI at IMO

Two cents on AI getting International Math Olympiad (IMO) Gold, from a mathematician. Background: Last year, Google DeepMind (GDM) got Silver in IMO 2024. This year, OpenAI solved problems P1-P5 for IMO 2025 (but not P6), and this performance corresponds to Gold. (1/10)

No. Ignore this. You are capable of reading books again. The brain is highly malleable and you only need a few screen-free weeks to recover your focus. You will be amazed how quickly the hunger for reading returns. You will be inhaling words

1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad (IMO).

I wrote an entry on Transformers for the Open Encyclopedia of Cognitive Science (@oecs_bot). I had to work with a tight word limit, but I hope it's useful as a short introduction for students and researchers who don't work on machine learning:

I'll briefly touch on arxiv.org/abs/2502.09863 (with Dhruva, Jamie, and Michael) and then discuss arxiv.org/abs/2505.18651 (with Daniel, Dhruva, and Matthieu).

General relativity 🤝 neural fields This simulation of a black hole is coming from our neural networks 🚀 We introduce Einstein Fields, a compact NN representation for 4D numerical relativity. EinFields are designed to handle the tensorial properties of GR and its derivatives.

Can an AI model predict perfectly and still have a terrible world model? What would that even mean? Our new ICML paper formalizes these questions One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵

Very happy to see my first "formal" work in theory of ML is now published, all thanks to awesome collaborators and mentorship of Prof. Cengiz Pehlevan! Learning about a field from scratch was a humbling experience -- I'm glad it shaped up into something uber cool!

Anthropic was at $1B in annualized revenue at the beginning of the year and is $4B now in mid-2025. OpenAI is at $10B. This kind of revenue growth has never been seen before by any company in human history.

[LG] Predictive representations: building blocks of intelligence arxiv.org/abs/2402.06590 This paper integrates reinforcement learning theory with cognitive science and neuroscience, focusing on the successor representation (SR) and its generalizations. It suggests that…

Exciting new preprint from the lab: “Adopting a human developmental visual diet yields robust, shape-based AI vision”. A most wonderful case where brain inspiration massively improved AI solutions. Work with @lu_zejin @martisamuser and Radoslaw Cichy arxiv.org/abs/2507.03168

it's strange that not many people know about this I randomly stumbled across -> aikosh.indiaai.gov.in looks like the @OfficialINDIAai is building "Huggingface" for datasets, models based on indian usecases seems like a really cool initiative. will try to contribute few…

A gentle reminder that TMLR is a great journal that allows you to submit your papers when they are ready rather than rushing to meet conference deadlines. The review process is fast, there are no artificial acceptance rates, and you have more space to present your ideas in the…