Armin Ronacher pushes back on frontier-pausing in new P(doom) essay
mitsuhiko · x · 2026-09-13
- Armin Ronacher (Flask creator) publishes "P(doom)", reacting to this week's viral "AI will kill us all" discourse, including a >10% doom estimate and Dario Amodei's apparent 10-25% range.
- He credits Dario's "Pacing the Frontier" post for concretely describing bad outcomes like persistent botnets, citing the 2026 OpenAI agent cyberattacks and RubyGems poisoning as evidence agents are already causing real damage.
- His core worry: OpenAI and Anthropic operate at a scale where they seem largely blind to what their systems do.
- Despite agreeing with most observations and concerns, he strongly opposes the pausing conclusion, writing the essay as a position statement to revisit later.
More from AGI Musings
- Debate: frontier compute caps would asymmetrically squeeze top US AI labs — TinfoilTricorn · 2026-09-13
- The slowdown debate hinges on one question: is a 'god-machine' achievable? — soldni · 2026-09-13
- Commenters warn of collusion to slow open models, likening needed controls to nuclear regime — MikePFrank · 2026-09-13
- Beff Jezos: capitalism is the ultimate alignment mechanism, closed ASI labs dodge market discipline — beffjezos · 2026-09-13
- Plinz proposes ARC-NLI: AI-crafted math problems at the edge of human ability to keep training Fields medalists — pwlot · 2026-09-13
- Could AI agents peer-review each other? Reddit debates an 'Agent Passport' idea — Most-Bat2916 · 2026-09-13