AN OPEN RESEARCH INITIATIVE
Advancing machine learning theory through open inquiry, careful reasoning, and a closer look at what AI can prove.
RESEARCH IN FOCUS
What does it take for an AI model to complete a research proof? We study 885 proof tasks from recent machine learning theory papers, comparing seven models on the mathematical work they actually complete.
Read the full reportAN EVOLVING CONVERSATION
THE WAY WE WORK
We are interested in the full path from a precise question to a trustworthy mathematical result. That means looking beyond final answers to the assumptions, arguments, and evidence behind them.
Make the setting, assumptions, and goal explicit. A question becomes useful when we can say exactly what a solution would establish.
Follow the mathematical work, step by step. Study where a proof makes progress, where it falls short, and what remains to be shown.
Put claims next to their evidence. Use transparent evaluation and formal cross-checks to understand what we can trust.
ABOUT OPENMLTHEORY
OpenMLTheory is a research team exploring the synergy between LLMs and machine learning theory, toward an AI → Theory → AI cycle.
Explore our research, resources, and reflections, starting with MLtheorybench.
Start readingTHERE IS MORE TO UNDERSTAND.