Portrait of Nikhil Prakash, Ascent Fellow 2027
2027 ASCENT FELLOW

Nikhil Prakash

PhD Candidate
Northeastern University
Trustworthy AI (Interpretability)

Highlighted research

Language Models Use Lookbacks to Track Beliefs

ICLR 2026

Biography

Nikhil Prakash is a fifth-year Ph.D. candidate in Computer Science at Northeastern University, advised by Prof. David Bau. His research focuses on demystifying the internal computations of deep neural networks. He has investigated how these networks support fundamental abilities such as variable binding, belief tracking and theory of mind, and spatial reasoning in vision-language models (VLMs). His work has uncovered that large language models (LLMs) and VLMs use representations resembling C programming language pointers to bind variables, including spatial variables. These models not only construct such representations but also manipulate them to perform complex reasoning. Furthermore, his research has found that fine-tuning LLMs primarily enhances mechanisms already learned by the base model rather than fundamentally introducing entirely new ones. These contributions have appeared at leading machine learning conferences, including ICLR, ICML, NeurIPS, and EMNLP, and have received recognition through spotlight presentations at workshops. Previously, he interned at Apple and Goodfire AI, investigating the internal mechanisms of LLMs.