A apolloresearch.ai More capable models are better at in-context scheming 6 points miles 1 year ago 1 comments
