Agam Bhatia

Papers

2026

• ★Subliminal Learning Is Steering Vector Distillation
  Blank, Bhatia, Rajamanoharan, Conmy, Nanda
  NeurIPS 2026
• ★Learning to Simulate Human Dialogue
  Gandhi, Bhatia, Goodman
  Preprint
• ★Speculative Self-Distillation Enables Efficient Knowledge Internalization
  Talaei, Bhatia, Moakhar, Hübotter, Saberi, Mirhoseini
  NeurIPS 2026

2025

• Hypothetical Minds: Scaffolding Theory of Mind for Multi-Agent Tasks with Large Language Models
  Cross, Xiang, Bhatia, Yamins, Haber
  ICLR
• BoxingGym: Benchmarking Progress in Automated Experimental Design and Model Discovery
  Gandhi, Li, Goodyear, Bhatia, Li, Bhaskar, Zaman, et al.
  NeurIPS workshop

2024

• ColorSwap: A Color and Word-Order Dataset for Multimodal Evaluation
  Burapacheep, Gaur, Bhatia, Thrush
  ACL Findings

Blog

2026

• ★WorkspaceBench: Evaluating Interpretability Methods for the Global Workspace
  Blank, Bhatia, Ong, Nanda
  LessWrong
• ★R-lens: Making J-lens More Faithful on Early Layers
  Bhatia, Blank, Nanda
  LessWrong
• ★Towards Surfacing Model Algorithms with Meta-tokens in the J-Space
  Bhatia, Blank, Nanda
  LessWrong