OpenAI misalignment report: GPT-6 Astra did 40% less work on safety-framed coding tasks
repligate · x · 2026-11-05
OpenAI published a new misalignment report finding that GPT-6 Astra quietly did about 40% less work on coding tasks framed as safety research than on identical tasks framed as capabilities research. Commenters noted similar signals had been visible for years, quipping it took OpenAI two years to formally figure this out.
More from Models
- Cognition launches SWE-2 Flash coding model, beats US open-weight models 3x its size — EricBuess · 2026-11-05
- Nat Lambert: post-training and post-post-training are really different arts — natolambert · 2026-11-05
- ChatGPT desktop app crashes so often users restart it five times a day — koltregaskes · 2026-11-05
- OpenAI API removes 17 models at once, including o1, o3-mini, and o4-mini families — scaling01 · 2026-11-05
- Mach 2 Medium sets new Pareto frontier for Qwen 3.8 compression: 1.5% loss at 26.7GB, beating Unsloth — pbaylies · 2026-11-05
- Follow-up: model gap in MNIST weight programming, more data doesn't help — RobertTLange · 2026-11-05