Agastya Agrawal

Welcome to my universe

Hi! I am Agastya Agrawal, a second-year undergraduate at the Chennai Mathematical Institute, pursuing my bachelor's in mathematics and computer science. I am passionate about interpreting artificial intelligence to help drive progress toward safe superintelligence. I particularly enjoy analyzing neural networks through a rigorous mathematical lens, and I am interested in applying mathematics, probability, and statistics to understand complex real-world structures. Please head to my blog or research section to learn more!

I am always open to collaborations or discussions about these topics, so please don't hesitate to reach out!

More about me →
Agastya Agrawal

Research

I study Singular Learning Theory, and in particular the connections it reveals within Reinforcement Learning. I am developing my understanding in a growing set of notes on Singular Learning Theory, which build up from its Bayesian foundations toward these ideas.

I also research deceptive models that exhibit unfaithful Chain of Thought. This builds directly upon my recent independent research, Uncovering Unfaithful CoT in Deceptive Models.

This June, I attended the Iliad Intensive, a month-long, mathematically rigorous course on foundational AI alignment, hosted at the London Initiative for Safe AI (LISA).

More on my Research →

Blog

This is the most interesting part of my universe! I document my notes here and share the cool mathematics and subjects that get me excited. Do check it out!

Read my blog →

Talks

I enjoy talking about mathematics and its applications, AI interpretability, and machine learning in general. Currently, I am hosting a Deep Learning Reading Group with TWIML.

TWIML Reading Group

Deep Learning Reading Group - Session 1

HOSTED BY TWIML

The kickoff session for our deep learning reading group. In this first talk, we cover the foundational concepts and set the stage for our upcoming deep-dives into models and alignment.

▶ Watch Recording Passcode: 9BAqrU6+
View all my talks →