Unverified50% confidenceFactExact time
Organizations such as OpenAI and Anthropic are working on "honesty" and "known unknowns," attempting to train models to proactively express uncertainty when uncertain through alignment techniques such as RLHF.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/12/2026
First Seen
Valid until: 12/11/2026
Sources
Related Entities
Related Claims
UnverifiedLeading organizations such as OpenAI and Anthropic both list calibrated uncertainty as a core metric for model safety and trustworthiness.73% similarUnverifiedAs companies such as OpenAI and Anthropic deepen their research into model alignment, the new generation of models is beginning to be trained with a 'helpful resistance' capability, pushing back when users reason incorrectly or make harmful requests.69% similarUnverifiedRLHF中人类评审倾向给流畅自信全面的答案更高分而较少关注事实核查，导致模型获得隐性激励表现得有把握比实际有把握更重要65% similarUnverifiedOpenAI's Preparedness Framework is essentially a self-assessment system; the evaluation methodology and specific test cases have not been fully disclosed, and there is no third-party independent verification mechanism.65% similarUnverifiedThe Poisson process is commonly used to model 'event arrival' problems, but requires independence and stationarity assumptions to be satisfied, whereas OpenAI's reset decisions are human-driven and do not fully satisfy these assumptions.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/912210API
curl https://kongchang.com/api/v1/knowledge/claims/912210MCP
get_claim(id=912210)