Scaling Misuse Detection with Self-Assessment
Comments
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Summary
Anthropic's new technique, self-assessment, has Claude append a short safety evaluation to each of its own conversations right after responding, reusing the cached context so the added cost is under 2% of inference compute.