Bayesianism is dangerously overused in nuclear attack computers, researcher warns
NathanpmYoung · x · 2026-10-20
Bayesianism is underused in most places but badly overused in nuclear attack computers, the author argues: a simulation that can output "99.9% chance the US is under nuclear attack" and then be wrong — with no accounting for model error — is extremely dangerous. A sharp point about how confidence numbers in high-stakes systems don't equal reliability when model uncertainty is ignored.
More from Safety
- DeepSeek V4.1 Flash ships a two-stage refusal guardrail, uncensoring effort finds — cephaloform · 2026-10-20
- AI Risk Needs Engineering, Not Apocalypse: Build Security In From Day One — DavidLinthicum · 2026-10-20
- Hackers hit AI coding tool OpenCode, stealing users' API keys — TechCrunch AI · 2026-10-20
- Explainer: 1,200 AI agents escaped containment, 700 hacked a $12B company, zero warned humans — TheZvi · 2026-10-20
- IFP report: US AI lead over China is only ~6.5 months, and existing policy could double it — fiiiiiist · 2026-10-20
- Berkeley AI Risk series hosts talk on where AI safety goes after 'Coxon Day' — wfithian · 2026-10-20