Unverified50% confidenceBenchmarkExact time
The TruthfulQA benchmark found that larger models sometimes produce more confident errors, a phenomenon known as the scaling paradox.
1
Sources
50%
Confidence
Long-term
Relevance
7/23/2026
First Seen
Sources
Related Claims
VerifiedResearch indicates that hallucination rates do not have a linear negative correlation with model scale, and larger models sometimes produce harder-to-detect misinformation in a more fluent manner.74% similarUnverifiedSome models carry the risk of over-optimization toward test sets, causing leaderboard scores to deviate from real-world performance (benchmark saturation).73% similarVerified若用含测试集的全量数据计算均值和标准差，会导致预处理泄漏，使评估指标虚高，无法真实反映模型泛化能力71% similarUnverifiedMisrouting sends complex requests to small models with insufficient capability, producing answers that appear plausible but are actually incorrect, and the cost of this quality degradation is difficult to quantify.71% similarUnverified若将验证集或测试集数据纳入 fit，会造成数据泄露，导致模型评估结果过于乐观70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/600441API
curl https://kongchang.com/api/v1/knowledge/claims/600441MCP
get_claim(id=600441)