Definition
AI-assisted research describes scientific workflows where large language models contribute ideas, computations, or drafts under human expert oversight — distinct from fully autonomous automated-research or ai-scientist systems.
Key Points
-
2026-07: levent-alpoge + claude-fable-5 jacobian-conjecture counterexample (jacobian-conjecture-claude-fable-counterexample, ai-mathematical-discovery)
-
Human-in-the-loop verification remains essential for formal proofs and experimental validation
-
LLMs can accelerate numerical reproduction and suggest proof strategies
-
Transparency about AI contributions (including errors) is a best practice in peer-reviewed work
Recent Developments
- 2026-07-01: giorgio-parisi and collaborators used claude to prove the jamming identity a+b=1; human experts corrected initial proof errors before JSTAT publication (2026-07-03-parisi-claude-jamming-conjecture-proof)