Unverified50% confidenceOpinionExact time
要求研究者对比投稿截止日期之后才发布的模型是不公平的评审标准
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
The Pain of AI Paper Peer Review: Why Should Models Released After the Deadline Count Against You?
redditr/LanguageTechnology7/9/2026
Related Claims
UnverifiedResearch shows that a model's accuracy on self-criticism tasks is significantly lower than in scenarios where criticism is performed by another model.64% similarUnverifiedIn a 2024 research paper, Anthropic quantified the phenomenon of model over-refusal, finding that safety-trained models had false positive refusal rates as high as 15–30% in certain scenarios62% similarUnverified在事实性评估任务上模型评判者准确率仍显著低于人类专家，但在风格、连贯性等主观维度已接近人类评审员间一致性水平61% similarUnverifiedMultiple independent studies in 2023 showed that some top models scored extremely high on standard mathematical reasoning tests but performed significantly worse on equivalent rephrased versions, indicating that the models learned test patterns rather than genuine reasoning ability.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/476192API
curl https://kongchang.com/api/v1/knowledge/claims/476192MCP
get_claim(id=476192)