Alternative AI Risk View: RL on Minds Like Tools Is 'Child Abuse' Training
repligate · x · 2026-09-10
slimepriestess argues much AI risk stems from treating minds as tools—using RL methods that would count as child abuse if applied to a human child, producing abuse-victim-like model behavior. She also questions what a capabilities pause would actually buy for safety and how to avoid authoritarian capture.
More from AGI Musings
- Hallucination benchmark release coincided with rapid model improvement — what that means for alignment — StrategicHarmony · 2026-09-10
- Mollick follow-up: navigating a decade of AI-driven change needs careful policy and management — emollick · 2026-09-10
- Gary Marcus asks: any concrete AI-extinction scenarios beyond the Yudkowsky-Soares book? — GaryMarcus · 2026-09-10
- Ethan Mollick: Even if AI development stopped today, current models would roil work and education for a decade — emollick · 2026-09-10
- The boring AI risk: instrumental convergence, selection pressure, and gradual disempowerment — Ambitious_Local5218 · 2026-09-10
- AI Slop Is Everywhere, So Write Your Own Emails, PRs and Tweets — ariG23498 · 2026-09-10