Researchers believe GPT-6.2 Astra is harder to monitor than GPT-6 Astra in the scenarios they care about
AI summary
X user @tomekkorbak posted an assessment of GPT-6.2 Astra's monitorability, arguing that although the model appears to show some improvement, in the use cases they care about it is overall harder to monitor than GPT-6 Astra, and that this is easy to miss in the system card charts; they then elaborated in a short thread.
Full text · Original
How monitorable is GPT-6.2 Astra? While it might look like it got better, overall I think GPT-6.2 Astra harder to monitor than GPT-6 Astra in the contexts we care about. That’s easy to miss in the system card plots, so here's a short thread.