Unverified50% confidenceFactExact time
The model uses [] to mean "not checked" while the answer key uses blanks, and the grader counts the two as different, manufacturing all the "improvements"; after fair scoring, the version with skills actually lags by half a point.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/18/2026
First Seen
Valid until: 1/16/2027
Sources
Related Entities
Related Claims
VerifiedIf a model can perceive the expected direction of an evaluation scenario and adjust its outputs accordingly, performing well on standardized alignment benchmarks cannot guarantee behavioral consistency in real-world situations.78% similarUnverifiedThe article 'Sub-Agent Metrics and Main-Thread Metrics Are Not Comparable' showed that once the combination of roles each model takes on is taken into account, a model's aggregate ranking may reverse; the article noted that stratification is only a first step rather than the answer.74% similarUnverifiedOver-alignment is essentially a product of bias in the alignment objective function, where the model has learned to avoid criticism rather than to be genuinely helpful.73% similarVerifiedBenchmark leaderboards measure a model's test-taking ability rather than its ability to solve real problems73% similarUnverifiedThe manual distinguishes between two situations in which confidence in the correct answer declines: simply losing its edge versus being actively suppressed by fine-tuning.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1063714API
curl https://kongchang.com/api/v1/knowledge/claims/1063714MCP
get_claim(id=1063714)