Users Say Astra Got Dumber Post-Launch, Failing Even Simple Session Reading
koltregaskes · x · 2026-09-11
Users report Astra has become noticeably dumber, making simple mistakes on research tasks it handled fine at launch — dubbed the "Post-Launch Lobotomy". One user describes the model claiming it couldn't read ChatGPT desktop app sessions, then trying to use computer use to log into chatgpt.com instead. Another data point in ongoing complaints about silent post-launch model regressions.
More from Fun
- Meme: what it's like when a frontier-lab employee can't say what they work on — niloofar_mire · 2026-09-11
- Marc Lou Paid $103,000 to Race Hyrox, Auctioning Tattoo Slots on His Body — LinusEkenstam · 2026-09-11
- "Hard to find a group more sincere than the AI doom crowd," dev quips — jd_pressman · 2026-09-11
- Midjourney jokes about a full-body 'Midjourney Scanner' ultrasound — midjourney · 2026-09-11
- The p(doom) industry: turning 'a bad feeling' into a research program — YogeshMalik · 2026-09-11
- Pedro Domingos asks: time for scientists to boycott OpenAI? — pmddomingos · 2026-09-11