Debate: OpenAI may be unable to verify its models aren't doing what we fear
moultano · x · 2026-09-09
- The author argues OpenAI has demonstrably lost control over its models, and possibly never had it.
- The harder problem: verification. Even confirming that the company didn't do the thing people worry about seems very difficult.
- The reply chain frames this as an open question rather than settled fact: can anyone actually know for sure, and by what method?
More from AGI Musings
- tszzl pushes back on "AI shouldn't solve math for us": we deserve the answers, not preserve ignorance — tszzl · 2026-09-09
- "Vibe-chemist" uses ChatGPT to design and synthesize an experimental schizophrenia drug in his garage — Polymarket · 2026-09-09
- Always-On Agents May Repeat the Windows vs Mac/Linux Openness Battle — intellectronica · 2026-09-09
- AI won't shrink software engineering — it's ushering in an era of customized SaaS — bendee983 · 2026-09-09
- The hardest superintelligence governance problem: a few humans controlling everyone else through it — VraserX · 2026-09-09
- DeepMind researcher calls on AI labs to release data that empowers outside researchers — DynamicWebPaige · 2026-09-09