AI safety debaters clash: is misaligned superintelligence or misuse of weak AI the bigger risk?
JacquesThibs · x · 2026-09-11
In an X debate on AI safety, the author argues that a misaligned superintelligence would not be dumb enough to kill just 1,000 people and then leave itself open to 'realignment.' He suggests harms of that scale are more likely to come from misuse of weak AI — a problem that still doesn't push the field toward superalignment research.
More from AGI Musings
- Why 'x% chance AI kills us' claims benefit from a huge attention incentive — dbasch · 2026-09-12
- Ex-Cohere researcher slams mathematicians' AI theory, predicts it will 'rug-pull' — suchenzang · 2026-09-12
- Everyone waiting may explain why AI's real-world impact is smaller than expected — moultano · 2026-09-12
- Dwarkesh podcast: Schulman, Millidge and O'Neill debate recursive self-improvement timelines — Dwarkesh Podcast · 2026-09-12
- Is AI capability growth causing a deflationary spiral that inhibits AI diffusion? — moultano · 2026-09-12
- Why work at AI companies if you think they might kill everyone? $500k+ helps — NathanpmYoung · 2026-09-12