Caleb Biddulph comments on Untrusted advice for AI control: Short, strong advice significantly uplifts weak LLMs

[Caleb Biddulph]

That’s pretty interesting! I wonder if it’s been tried before in AI control.