More capable models can better recognize the specific circumstances under which they are trained. Because of this, they are more likely to learn to act as expected in precisely those circumstances while behaving competently but unexpectedly in others. This can surface in the form of problems that Perez et al. (2022) call sycophancy, where a model answers subjective questions in a way that flatters their user’s stated beliefs, and sandbagging, where models are more likely to endorse common misconceptions when their user appears to be less educated.
Recent articles
- Notes on the new Claude analysis JavaScript code execution tool - 24th October 2024
- Initial explorations of Anthropic's new Computer Use capability - 22nd October 2024
- Everything I built with Claude Artifacts this week - 21st October 2024