DeepMind's Lampinen: LMs may make 'intentional' discoveries of new math concepts
AndrewLampinen · x · 2026-10-03
Andrew Lampinen (Google DeepMind) argues that LMs sometimes exhibit more interesting forms of generalization enabling "intentional" discovery — a model might realize a cheat code likely exists and actively search for it. He suggests this could extend to blue-sky discovery, such as a model hypothesizing useful new mathematical abstractions the way mathematicians do. The challenge, he notes, is researching this: "there is no new thing under the sun," making genuine novelty hard to test.
Related event: DeepMind researcher suggests LLMs can actively discover new abstractions(2 posts)→
More from AGI Musings
- Chip design is a ~10^2,632,341 search problem — AI and agents are turning hardware into search — ai · 2026-10-03
- What happens to local accents when every message gets polished by AI? — yi111 · 2026-10-03
- Vibecoded cross-game 'mods' slammed as mere real-time compositing as natural-language coding era begins — cgarciae88 · 2026-10-03
- Reddit thread: what are your AI/ML predictions for 2027 and 2028? — imfamilyfriendlysd · 2026-10-03
- Ethan Mollick's new book Co-Existence gets a blurb written for AI readers — emollick · 2026-10-03
- AI engineering is like making law, not playing games: rules must shift as they meet reality — danshipper · 2026-10-03