Ex-Meta Researcher Fleuret Proposes Token-Tagging Approach to AI Alignment

Former Meta/Apple researcher François Fleuret proposed an alignment approach combining RL during pretraining with tagging every token as "heard, thought, or said," while critics questioned how such labels could resist forgery by malicious documents without independent verification.

2026-10-28 ~ 2026-10-28 · 3 related posts

• François Fleuret Proposes Tagging Every Token 'Heard/Thought/Said' to Solve Alignment — francoisfleuret · 2026-10-28
• Follower challenges token-tagging proposal: who stamps the tags to stop forged labels? — dl_insider · 2026-10-28
• Fleuret details token tagging for alignment: a second readout classifying heard/thought/said — francoisfleuret · 2026-10-28