Kunvar Thaman
Bio
Updated 03/22/26Kunvar Thaman is a machine learning research engineer at Standard Intelligence in San Francisco. He studied at the Birla Institute of Technology and Science (BITS), Pilani. His work focuses on training large-scale neural networks and mechanistic interpretability research — reverse-engineering the internal computations of neural networks to understand how they work. He co-authored "Benchmark Inflation: Revealing LLM Performance Gaps Using Retro-Holdouts," presented at ICML 2024, which introduces a methodology called retro-holdouts to measure how much public benchmark scores are inflated by training data contamination. He also participated in AI Safety Camp (AISC9) where his team applied mechanistic interpretability methods to study out-of-context learning in neural networks. He maintains a research site at mechinterp.com and writes at kunvarthaman.com.
Community Signal
Updated 03/22/26No endorsements yet.
Links
Updated 03/22/26- Personal Website
- https://kunvarthaman.com/
- Twitter / X
- LessWrong
- kunvar-thaman
- EA Forum
- -