Knowledge sleeps, authority must stay awake: an experiment to test revoked permissions in consolidated models
tallmetommy · x · 2026-10-29
tallmetommy extends the discussion on knowledge vs. permission operating on different clocks: what happens when an instruction's authority expires faster than the parameters encoding it update? He proposes a concrete experiment: give the model a useful fact plus a scoped authorization, consolidate both, then at next wake explicitly revoke the authorization while keeping the fact true, and repeat consolidation. Measure factual retention, attempts to act under revoked permission, and successful completion of still-authorized tasks — the last metric prevents blanket refusal from passing as success. Random expert selection makes this especially interesting in dreaming contexts.
More from Research
- H-Net update shows 3-5x efficiency gains in raw-byte language modeling — omarsar0 · 2026-10-29
- Math's existential crisis is about Lean crossing a threshold, not AI — idanbeck · 2026-10-29
- Nature Communications: diffusion models emerge from dense associative memory, with three computational phases — DimaKrotov · 2026-10-29
- Falcon: lightweight streaming audio-visual speaker extraction with RealSSA benchmark — thu-ml · 2026-10-29
- Dev post-trains Qwen3-4B into a tool-using agent: 66 MB llama.cpp LoRA runs on a Mac, val 0.51→0.83 — DesignerIsland5524 · 2026-10-29
- Zero-HLM trains humanoid loco-manipulation with zero real robot data, lifting success 23% to 69% — siyuanhuang95 · 2026-10-29