Karthik R. Narasimhan

Karthik R. Narasimhan is a Professor at Princeton University's School of Engineering and Applied Science in the Department of Computer Science. Previously, he earned his PhD from MIT under Regina Barzilay and served as a visiting research scientist at OpenAI during 2017-18. His research focuses on the intersection of language and decision-making, building autonomous agents that learn from both experience and human knowledge.

His research spans multiple high-impact areas including language agents (Text-DQN, CALM, ReAct, Tree of Thoughts), reinforcement learning (h-DQN, Multi-Objective RL), and AI safety (Toxicity in ChatGPT, DataMUX). He has developed critical datasets and benchmarks such as WebShop, InterCode, SWE-bench, and SILG that have become standard evaluation tools in the field. Current work emphasizes agent capabilities, software engineering automation, and multimodal interaction.

His publication trends show strong focus on practical agent deployment (SWE-agent, Tree of Thoughts), safety evaluation (Probing AI Safety), and efficiency improvements (DataMUX). Recent work increasingly addresses real-world challenges in software engineering, security, and human-AI collaboration through rigorous benchmarking.

• Co-author of foundational GPT (2018) paper
• Key developer of Text-DQN (2015), CALM (2020), ReAct (2022), Tree of Thoughts (2023)
• Creator of influential benchmarks: WebShop (2022), SWE-bench (2023), InterCode (2023)

He actively advises students through Princeton's computer science program, with research supported by multiple grants focused on autonomous agent development and language-based decision systems. His GitHub repositories (nlp-datasets, text-world-player) demonstrate strong community engagement in open-source research tools. Current projects include advancing language agent capabilities through Reflexion (2023) and Tree of Thoughts (2023) frameworks while addressing critical safety and efficiency challenges.