A new colloquium by the Institute for AI & Fundamental Interactions (IAIFI) explores "LLM Processes," a framework that enables scientists to incorporate expert domain knowledge into machine learning models using plain language, rather than complex code.
The talk, streamed live on April 24, 2026, presents a method that treats large language models (LLMs) as flexible priors for scientific modeling. By describing physical principles, experimental constraints, or known relationships in natural language, researchers can guide an LLM to generate plausible hypotheses or parameter distributions. This approach aims to bridge the gap between human intuition and data-driven AI, making scientific modeling more accessible and grounded in established knowledge.
IAIFI, a research center focused on the intersection of AI and fundamental physics, regularly hosts colloquia on cutting-edge topics. This session highlights how LLMs can serve not just as chatbots, but as collaborative tools for hypothesis generation and model specification in the sciences.