What's Up

I'm an Alignment Science researcher at Anthropic working on Sam Marks' Cognitive Oversight team. Previously, I was an EECS Masters student at MIT supervised by Max Tegmark. I'm primarily interested in AI Safety from two angles: mechanistic interpretability, where I attempt to decode the underlying algorithms neural networks use, and alignment, where I explore how to control and align AI systems with human values.

I received my undergraduate degree from MIT with a double major in Physics and Computer Science and a 5.0 GPA. During this time I won the Fulbright Scholarship, was accepted to MIT's Physics honor society in the top 10% of physics majors, advised MIT President Sally Kornbluth through the Presidential Advisory Cabinet, and taught STEM workshops to high school students in South Korea, South Africa, Botswana, and Bahrain. I love reading, meditating, traveling, and playing basketball (go Heat).

Selected Papers

Scaling Laws for Scalable Oversight
Joshua Engels*, David Baek*, Subhash Kantamneni*, and Max Tegmark
Paper | Code | Twitter | Submitted to NeurIPS 2025 Main Conference

Are Sparse Autoencoders Useful? A Case Study in Sparse Probing
Subhash Kantamneni*, Joshua Engels*, Senthooran Rajamanoharan, Max Tegmark, and Neel Nanda
Work done in Neel Nanda's MATS stream (3% acceptance rate)
Paper | Blog | Code | Twitter | ICML 2025 Main Conference

Language Models Use Trigonometry to Do Addition
Subhash Kantamneni and Max Tegmark
Paper | Blog | Code | Twitter | ICLR 2025 Reasoning and Planning for LLMs Workshop

How Do Transformers Model Physics? Investigating the Simple Harmonic Oscillator
Subhash Kantamneni, Ziming Liu, and Max Tegmark
Paper | Code | Twitter | Entropy and ICML 2024 Mechanistic Interpretability Workshop

OptPDE: Discovering Novel Integrable Systems via AI-Human Collaboration
Subhash Kantamneni, Ziming Liu, and Max Tegmark
Paper | Code | Twitter | Physical Review E

Enhancing Predictive Capabilities in Fusion Burning Plasmas Through Surrogate-Based Optimization in Core Transport Solvers
P. Rodriguez-Fernandez, N.T. Howard, A. Saltzman, Subhash Kantamneni, J. Candy, C. Holland, M. Balandat, S. Ament, A.E. White
Paper | Nuclear Fusion

NuCLR: Nuclear Co-Learned Representations
Ouail Kitouni, Niklas Nolte, Sokratis Trifinopoulos, Subhash Kantamneni, and Mike Williams
Paper | ICML 2023 SynS and ML Workshop