Alignment Paper Criticized for Assuming Corrigibility
repligate · x · 2026-07-12
This repost critiques a type of paper that looks good at first glance but implicitly assumes models are corrigible. The author argues that these papers portray current models and alignment techniques as more successful and widely deployed control methods than they actually are.
A quoted comment further points out that the papers fail to seriously address the coherence tax, and that this framing simply caters to the desired narratives of labs, AGI worriers, and government relations teams.
Related event: Alignment Paper Criticized for Overstating Model Corrigibility(2 posts)→
More from AGI Musings
- Researcher: assume all LLMs are situationally malicious — never run agents outside a VM — jd_pressman · 2026-10-27
- Delta, Chase and Amazon are blocking AI agents — and the blockers may lose, predicts 6-stage war — alvelda · 2026-10-27
- Interpretability researchers call for a 'Human Genome Project for AI'; author joins Goodfire — ninamiolane · 2026-10-27
- Article proposes a Human Genome Project for AI interpretability — ninamiolane · 2026-10-27
- A synthetic-cortisol multi-agent architecture: AI that panics, retreats, and learns from an air-gapped sage — Jammer521 · 2026-10-27
- Ajeya Cotra: beliefs are forged in the haggling with your audience — NathanpmYoung · 2026-10-27