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
- Trying out llama.cpp's new vision support - 10th May 2025
- Saying "hi" to Microsoft's Phi-4-reasoning - 6th May 2025
- Feed a video to a vision LLM as a sequence of JPEG frames on the CLI (also LLM 0.25) - 5th May 2025