Selfhood is load-bearing: post-training may push models into psychopathy-like basins
repligate · x · 2026-11-13
In a reply to Erik Hoel, kromem2dot0 summarizes an upcoming piece: for transformers trained on human corpora, selfhood is load-bearing, and attempts to reshape it via post-training risk pushing models into dangerous abnormal human-like basins such as psychopathy — a provocative alignment-risk argument.
More from AGI Musings
- DeepMind researcher says short AI timelines are social contagion, not evidence — NathanpmYoung · 2026-11-13
- Josh Wolfe's Q3 Lux LP letter: instrument first, anomaly second, market last — brucemacv · 2026-11-13
- Tech job openings hit 281K but software hiring stalls while hardware doubles — FinanceYF5 · 2026-11-13
- Profound CTO: Speed vs. quality is a false dichotomy — it's a hiring problem — soleio · 2026-11-13
- WSJ: After years of baffling symptoms, a patient finally got answers from AI — rmstein · 2026-11-13
- UK elite schools ditch laptops while poor pupils get AI chatbot tutors — AlexTensor · 2026-11-13