Kanishk Gandhi
I am interested in building machines that understand people.
Previously: PhD, Stanford (with Noah Goodman) · Brenden Lake & Moira Dillon, NYU · Microsoft Research (with Dimitris Papailiopoulos) · PathAI · IIT Kanpur
Publications
2026
-
Building Machines that Understand People
— PhD thesis, Stanford University
Thesis -
QuasiMoTTo: Quasi-Monte Carlo Test-Time Scaling
— Preprint 2026
arXiv -
Learning Next Action Predictors from Human-Computer Interaction
— CoLM 2026
arXiv -
Endless Terminals: Scaling RL Environments for Terminal Agents
— Preprint 2026
arXiv -
Learning to Simulate Human Dialogue
— Preprint 2026
arXiv
2025
-
Not All Thoughts Matter: Selective Attention for Efficient Reasoning
— NeurIPS 2025 Workshop on Efficient Reasoning
OpenReview -
Scaling up the think-aloud method
— CogSci 2025 (Oral)
arXiv -
Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs
— CoLM 2025 (Oral)
arXiv Media: CNBC -
Big-Math: A Large-Scale, High-Quality Math Dataset for Reinforcement Learning in Language Models
— Preprint 2025
arXiv -
D3: A Dataset for Training Code LMs to Act Diff-by-Diff
— CoLM 2025
OpenReview -
Non-literal Understanding of Number Words by Language Models
— CogSci 2025
arXiv -
Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought
— Preprint 2025
arXiv -
BoxingGym: Benchmarking Progress in Automated Experimental Design and Model Discovery
— NeurIPS 2025 — Scaling Environments for Agents Workshop
arXiv
2024
-
Stream of Search (SoS): Learning to Search in Language
— CoLM 2024 (Oral)
arXiv -
Procedural Dilemma Generation for Evaluating Moral Reasoning in Humans and Language Models
— CogSci 2024 (Oral) , NeurIPS Workshop 2023
arXiv NeurIPS Workshop -
Surveying the Effects of Quality, Diversity, and Complexity in Synthetic Data from Large Language Models
— Preprint 2024
arXiv -
Psychometric Alignment: Capturing Human Knowledge Distributions via Language Models
— Preprint 2024
arXiv -
Self-Supervised Alignment with Mutual Information: Learning to Follow Principles without Preference Labels
— NeurIPS 2024
arXiv -
Human-like Affective Cognition in Foundation Models
— Preprint 2024
arXiv
2023
-
Social Contract AI: Aligning AI Assistants with Implicit Group Norms
— NeurIPS Socially Responsible Language Modeling Research Workshop 2023 (Oral)
arXiv -
Understanding social reasoning in language models with language models
— NeurIPS 2023 (Spotlight)
arXiv Code -
Certified Deductive Reasoning with Language Models
— TMLR 2023 ICLR 2025 (Presentation)
arXiv Code -
Strategic Reasoning with Language Models
— NeurIPS Foundation Models and Decision Making Workshop 2023
arXiv -
Intuitions about physical scenes and objects in Virtual Reality (VR)
— CogSci
-
Commonsense Psychology in Human Infants and Machines
— Cognition 2023
Media: NSF Science Now WNYC/Gothamist The Daily Beast The Jerusalem Post NYU News Washington Square News ScienceDaily
2022
-
Eliciting Compatible Demonstrations for Multi-Human Imitation Learning
— CoRL 2022
2021
-
Baby Intuitions Benchmark (BIB): Discerning the goals, preferences, and actions of others
— NeurIPS 2021
-
Evaluating infants' reasoning about agents using the Baby Intuitions Benchmark (BIB)
— CogSci 2021
2020
-
Mutual exclusivity as a challenge for deep neural networks
— NeurIPS 2020
Media: New Scientist