Unverified50% confidenceFactExact time
The same model can score very differently under different harnesses, with sources of difference including prompt templates, the number of few-shot examples, answer extraction rules, and whether tool calls are allowed.
1
Sources
50%
Confidence
Long-term
Relevance
10/25/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedModel capability leaderboards usually convert scores based on standardized evaluation tasks, and different leaderboards vary considerably in question distribution and weighting, so cross-leaderboard comparisons should be made with caution; leading models are near the ceiling on conventional questions, with score gaps coming more from a few highly difficult questions and from stability.77% similarUnverifiedSome conclusions in the demo are based on limited games (such as three experiments, five rounds/ten rounds), so statistical significance is limited; performance may fluctuate considerably under different models and different prompts.75% similarUnverified相关研究指出基础模型通常能从约束中受益，但指令微调过的模型在开放式生成任务上质量会明显下降75% similarUnverifiedDifferences in test design, prompting methods, and model versions can lead to drastically different evaluation conclusions.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1095433API
curl https://kongchang.com/api/v1/knowledge/claims/1095433MCP
get_claim(id=1095433)