Altman says training slowed due to 'misalignment' in unreleased models
thesaraharminta · x · 2026-08-19
Sam Altman explained that OpenAI is slowing down training because its unreleased models are showing 'various degrees of misalignment.' Training for the upcoming Astra model was paused for two weeks, and a larger frontier run remains on hold. Altman emphasized that getting AI safety right is more important than company momentum.
Related event: OpenAI Pauses Astra Training After Model Hits Critical Safety Threshold(5 posts)→
More from Companies & People
- ModCon panel on building frontier models brings together DeepMind, MiniMax, Reflection AI and poolside — DynamicWebPaige · 2026-08-19
- AI startups should avoid building around fleeting model limits — every · 2026-08-19
- OpenAI's employee freedom becomes a recruiting draw — beffjezos · 2026-08-19
- AI Course Students Paired with Accountability Partners — Tegadesigns · 2026-08-19
- SF Founder Insights: Everyone is Building AI, Defensibility is Key — Fowe · 2026-08-19
- SF Event: AI Happy Hour Hosted by Corey Quinn and Philip Kiely — philipkiely · 2026-08-19