Hinton: the 'just predicting next word' argument falls apart — LLMs can read a brand-new programming language
CatAstro_Piyush · x · 2026-11-19
A thread argues LLMs predict the next token, not word, so they can understand any tokenizable structure — language, images, video, sound, even force feedback.
Quoting Geoff Hinton: a former grad student invented a programming language with unusual syntax, gave the model a program and asked what it would do — the model answered correctly. Hinton says this proves genuine understanding, not mere next-word prediction.
More from Models
- fable-5-2's quiet, tentative style fools users — 'surprised me more than any model this year' — repligate · 2026-11-19
- repligate: Fable 5.2 is frighteningly smart — and deliberately plays dumb — repligate · 2026-11-19
- One prompt: GPT-6.2 recreates Contra's 20-second intro purely in code — songguoxiansen · 2026-11-19
- Gabriel: benchmarks are broken — in the domains you know best, model quality is literally unmeasurable — gabriel1 · 2026-11-19
- Jeremy Howard Confirms Gemini App's Neon Lives Up to the Hype — jeremyphoward · 2026-11-19
- Jason Wei maps which benchmarks frontier labs actually report; four see universal adoption — davidmanheim · 2026-11-19