Study: LLMs Can Infer Causal Structure From Distributional Semantics Alone
AndrewLampinen · x · 2026-09-28
AI researcher Andrew Lampinen debates whether the lack of meaning in language form implies a system cannot infer underlying causal structures — the classic octopus thought experiment. He argues a new study shows a system can indeed learn sufficient causal structure from distributional semantics and generalize well even when assessed in language alone.
More from AGI Musings
- Today's scary internal frontier model is tomorrow's toy users complain about — nptacek · 2026-09-28
- Ex-Anthropic safety researcher: racing to RSI is hubris, not a prisoner's dilemma — dgrobinson · 2026-09-28
- Yann LeCun: Scaling LLMs alone will 'absolutely never' reach AGI — MickeySteamboat · 2026-09-28
- Dev: drop the AGI chatter until models get continual learning — frozen weights will never be general enough — cgarciae88 · 2026-09-28
- Microsoft agent ports Copilot runtime to Rust in 25 hours for $120K, per LifeArchitect Memo — dralandthompson · 2026-09-28
- Scientist: AI using ideas isn't the problem — missing attribution is — arjunrajlab · 2026-09-28