Unverified50% confidenceBenchmarkExact time
2024 年的多项基准测试表明，经过严格安全对齐的模型在代码生成任务上的拒绝率比基座模型高出 15-30%，在创意写作任务上的自我审查率高达 40%
1
Sources
50%
Confidence
Long-term
Relevance
8/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn 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 scenarios74% similarUnverifiedResearch shows that high-quality prompts can bring a 20%–40% improvement in task completion rates compared to casually written prompts under the same model conditions.73% similarUnverified推测解码中当草稿通过率低于50%时，实际加速效果会大打折扣甚至产生负优化73% similarUnverifiedA 2023 study by Stanford University showed that Copilot's error rate in security-sensitive code scenarios is as high as 40%.72% similarUnverifiedResearch shows that structured prompts, compared to casual descriptions, can improve the first-pass rate of code generation from 30% to over 70%71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/786981API
curl https://kongchang.com/api/v1/knowledge/claims/786981MCP
get_claim(id=786981)