Unverified50% confidenceBenchmarkExact time
In the 30-day holdout evaluation, untuned HistGBR (B5) approached TabPFN with an MAE of 10.8 on the validation month, and TabPFN's ranking advantage over B2 was not significant at that sample size.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the 30-day holdout evaluation, TabPFN v2 had the lowest error with an MAE of 10.3 and RMSE of 14.6, and the 95% confidence interval for its MAE advantage over B2 was −17.0 to −6.9.72% similarUnverifiedUsing a 4B model to rerank the candidate shortlist filtered by BM25 outperformed BM25 alone (+0.047, confidence interval [0.020, 0.073]), even though the 4B reranker itself did not pass qualification.63% similarUnverifiedIn pre-deployment evaluations of Claude Sonnet 4.5, the model achieved a 0% misalignment rate, but reading its chain of thought revealed that it clearly knew it was being tested.62% similarUnverified在图片隐藏虫子识别测试中，Sonnet 5的Max模式花费约10分钟并识别结果完全正确，而Gemini顶配模型只找到4只，GPT-5.5 X-High未能检测出60% similarUnverifiedIn the LongMemEval benchmark, BM25 accuracy improved from 78.8% to 85.2% with 5 chunks.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1002628API
curl https://kongchang.com/api/v1/knowledge/claims/1002628MCP
get_claim(id=1002628)