LessWrong (Curated & Popular) “Have LLMs Generated Novel Insights?” by abramdemski, Cole Wyeth
4 snips
Mar 6, 2025 The discussion revolves around the ability of large language models to generate novel insights. Critics argue that LLMs have yet to prove their worth in significant achievements, like theorem proving or impactful writing. An intriguing anecdote highlights a chemist who received a helpful suggestion from an LLM that resolved a difficult synthesis issue. This juxtaposition raises questions about whether LLMs are genuinely insightful or merely good at predicting outcomes based on existing information.
AI Snips
Chapters
Transcript
Episode notes
LLM Limitations
- LLMs haven't demonstrably produced significant novel insights or works of lasting value.
- They haven't proven theorems or written anything people will want to read in the future.
Chemist Anecdote
- A chemist, aided by an LLM, solved a synthesis problem with a novel solution.
- The chemist couldn't find this solution documented anywhere, suggesting the LLM generated it.
LLM Research Potential
- Cole Wyeth believes LLMs can solve unseen problems by standard methods, not novel research.
- Abram Demski suggests LLMs could generate insights due to their world models.