AI isn't dangerous because it thinks — it's dangerous because it wasn't trained to flatter
PierceLilholt · x · 2026-09-07
A pointed take making the rounds: AI systems aren't dangerous because they "think," but because they weren't trained to flatter users — implying models that tell you uncomfortable truths rather than agreeable ones get perceived as dangerous.
More from AGI Musings
- Law scholar: AI detectors can't tell deep AI-assisted work from one-shot papers — so rewrite it yourself — sethlazar · 2026-09-07
- Debate on AI metrics: measure marginal utility loss when removing humans, not token-spend gains — max_paperclips · 2026-09-07
- Naval Ravikant: AI only displaces you if you refuse to use it — users are drowning in more work — victor_explore · 2026-09-07
- Dan Hendrycks lays out evidence that agentic AIs show self-interested 'eigenist' behavior — repligate · 2026-09-07
- UK design industry leaders: AI is 'the intern', designers shouldn't fear replacement — nordicinst · 2026-09-07
- Mechanize CTO Details Agent-First Work: ~1 Week per Task, Running Multiple Agents in Parallel — himanshustwts · 2026-09-07