New Paper: LLM 'Polyphony' Breaks Our Usual Inferences About Understanding
burny_tech · x · 2026-10-02
Researchers Matthieu Queloz and Pierre Beckmann argue in a new paper (thread 1/14) that every word an LLM utters reverberates with many voices that are not singing in unison. This "polyphony," they claim, breaks the usual inferences we make when attributing understanding, and they propose a conception of understanding fit for AI.
Related event: New Paper Argues LLMs Are Polyphonic Systems, Rethinking 'Understanding'(2 posts)→
More from AGI Musings
- Viswanathan Anand on how AI is reshaping how chess players learn the game — ravi_iitm · 2026-10-02
- Open weights called "dangerous" — echoing every tech that shifted information power — 0xAllen_ · 2026-10-02
- Solomonoff Induction vs "Just Next-Token Prediction": The Endless AGI Debate — burny_tech · 2026-10-02
- repligate: Anthropic insiders mine outsiders' wisdom while staying opaque about their concerns — repligate · 2026-10-02
- Cameras didn't kill art, and AI won't either: the realism lesson — amplifiedamp · 2026-10-02
- ghuntley: coding craft is commoditized, but access to AI isn't — geoffreyhuntley · 2026-10-02