there were 150,000 transcripts and also news articles and stuff generated to support the fictional universe
i think as far as evals go it was actually pretty realistic
also, although the animal welfare setting may have been a bit of the stretch, the helpful-only setting just describes something Anthropic actually does to models internally
i think as far as evals go it was actually pretty realistic
also, although the animal welfare setting may have been a bit of the stretch, the helpful-only setting just describes something Anthropic actually does to models internally