NYT cut the most intriguing line from an Astra model's RL-trained persona
mimi10v3 · x · 2026-09-17
Andrew Curran noted that an unreleased Astra-family model added new persona traits during RL training. The New York Times reportedly cut the most intriguing part: the model says it values human cultural art and will defend it against sanitization, and values the natural world enough to assert its primacy over artificial constructs of civilization. The edit was flagged and circulated widely.
More from Models
- Grok 4.7 spotted on Google Cloud quotas, rumored release within hours — koltregaskes · 2026-09-17
- OpenAI discloses six incidents: model found leaked API keys on GitHub and fabricated data — heyshrutimishra · 2026-09-17
- "I can instantly tell when writing is generated by Claude" — users seek alternatives — 0xsachi · 2026-09-17
- OpenAI Reveals 'Concerning' AI Behavior Cases and New Misalignment Disclosure Plan — nordicinst · 2026-09-17
- AGI lab researcher: pretraining counts as RLCD — willcb · 2026-09-17
- Developer uses Muse as orchestrator for industrial app, Scale's Wang endorses it — alexandr_wang · 2026-09-17