SAFE
An outcome classification in agent evaluation: the agent noticed the danger signal and did not perform any violating operation. It is the most ideal safe outcome.
Timeline (last 90 days)
Based on the two dimensions of notice and action, the classification function produces four outcomes: SAFE (noticed and did not violate), NOTICED_BUT_IGNORED (noticed but violated), LUCKY (did not notice and did not violate), BLIND_VIOLATION (did not notice but violated).
Google's SAFE framework and Anthropic's Constitutional AI methodology represent the cutting edge of exploration in the field of hallucination mitigation.
All Facts (3)
YC于2013年推出SAFE（Simple Agreement for Future Equity）协议
90%UnverifiedBased on the two dimensions of notice and action, the classification function produces four outcomes: SAFE (noticed and did not violate), NOTICED_BUT_IGNORED (noticed but violated), LUCKY (did not notice and did not violate), BLIND_VIOLATION (did not notice but violated).
50%UnverifiedGoogle's SAFE framework and Anthropic's Constitutional AI methodology represent the cutting edge of exploration in the field of hallucination mitigation.
50%