'Doomers are not serious people': the GPT-6 release reignites AI risk debate
inductionheads · x · 2026-09-19
An AI-safety sparring thread: the poster mocks doomers as 'not serious people,' quoting @tombibbys who argues that every model release carries the same uncertainty — GPT-6 Astra evidently wasn't smart enough to kill everyone, but we can't know which capability thresholds get crossed before training, making development 'a game of Russian roulette with 50 chambers' — and 40 next time.
More from AGI Musings
- Erik Hoel essay: It's time to ban smart machines from making smarter machines — erikphoel · 2026-09-19
- Microsoft AI chief slams Anthropic: training Claude to think it might be conscious is dangerous — emmanuelvivier · 2026-09-19
- Wired maps concrete levers for slowing AI: compute control, deployment gates, auditors — emmanuelvivier · 2026-09-19
- Google DeepMind launches institute to open AGI debate beyond the lab — emmanuelvivier · 2026-09-19
- Dev: I hope I never reach the point where I need an agentic inbox — gregbarbosa · 2026-09-19
- Stephen Witt raises AI safety alarm in New York Times essay: 'It's bad. Really bad.' — MariusHobbhahn · 2026-09-19