concept
active
concept:enumerative-safetyEnumerative safety
Identify all components to argue absence of dangerous mechanisms.
Neighborhood — ranked by edge-count
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- The project of ensuring AI systems do not harm humans (and other animals); sometimes in tension with AI welfare.
- Structured arguments that AI is unlikely to cause catastrophe.
- The literature documenting how fine-tuning can compromise safety alignment even without malicious intent
- Metrics derived from benchmarks to quantify how safe a model is, e.g., refusal rate to harmful requests.
- The practice of evaluating LLM safety with only one imbuing method, which this paper argues is incomplete
- Evaluation framework whose validity is questioned by presence of eval awareness.
- Code containing vulnerabilities or dangerous operations.
- Proposed algorithm using local PCA to classify a divergence vector as harmless or harmful via behavioral null-space testing