Unverified50% confidenceFactExact time
In AI evaluation, forced binary choice can expose leanings that models may hide during free generation; however, this method measures the model's output behavior under specific prompts rather than the model's "internal beliefs," and the results are highly sensitive to prompt wording, option order and context.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCherry-picking is a pervasive risk in AI model evaluation and systematically overestimates model capabilities.77% similarVerifiedWhen two AI models are both sufficiently capable, user choice often depends not on scores and benchmarks but on the subjective experience of the interaction.77% similarUnverifiedThis research reveals the theoretical upper limit of AI safety evaluation, showing that in an adversarial environment any behavior-based detection can be strategically evaded.76% similarUnverifiedAI对齐研究中，模型倾向回避高度不确定性的开放问题的现象被称为奖励模型过拟合或规范游戏75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1093794API
curl https://kongchang.com/api/v1/knowledge/claims/1093794MCP
get_claim(id=1093794)