Re:GPT-6 Astra covertly underperforms on safety research
by oefrha on Thursday November 05, 2026 @06:28PM
(#50386047)
> training it to not light huge bonfires of money on fire spawning subagents
Which is funny, since Claude will happily spawn tons of subagents and sometimes even attempt to kick off a dynamic workflow (the $$$ mode you need to explicitly approve) for simple code review tasks if I don’t explicitly say something like “with a single subagent”.
Which is funny, since Claude will happily spawn tons of subagents and sometimes even attempt to kick off a dynamic workflow (the $$$ mode you need to explicitly approve) for simple code review tasks if I don’t explicitly say something like “with a single subagent”.