Automating the Search for Knowledge | NeurIPS 2025 скачать в хорошем качестве

Automating the Search for Knowledge | NeurIPS 2025

We are moving beyond Chatbots. In this interview from NeurIPS 2025, Bodhisattwa Prasad Majumder breaks down how "Scientific Agents" are being built to automate the research process itself. Bodhi's team is treating scientific discovery as a search problem—using agents to generate hypotheses, navigate massive datasets, and validate findings across 7 different domains, from economics to hydrogeology. We discuss the shift from passive LLMs to active agents and the "North Star" goal: an AI system capable of authoring a paper in a leading journal. Timestamps: 00:00 Intro: The Shift to Scientific Agents 00:37 Treating Science as a "Heuristic Search" Problem 02:01 The Allen Institute's Open Source Mission 03:24 How AI Improves Scientific Reproducibility 04:23 The "Tribal Knowledge" Problem in Research 05:52 Auto-Discovery: Testing Agents in 7 Fields 06:45 The North Star: AI as a Lead Author 07:28 The Future of AI at NeurIPS Thank you to ‪@lambda-ai‬ for sponsoring these NeurIPS interviews!