Unverified85% confidenceFactTime unknown
EleutherAI的Language Model Evaluation Harness专注于能力基准测试
1
Sources
85%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
Anthropic捐赠AI对齐工具Petri给Meridian Labs：开源安全评估新格局
twitterAnthropicAI
Related Entities
Related Claims
Unverified探测（probing）技术最早广泛应用于自然语言处理领域，用于检测语言模型是否编码了语法结构、语义角色等语言学特征，Alain和Bengio 2016年做了开创性工作70% similarUnverified评估大语言模型能力的常见基准测试包括MMLU、HumanEval、MATH以及LMSYS Chatbot Arena69% similarUnverified语言学习场景选购路径：先确定主学语种→查该语种是否支持离线识别与真人发音→确认是否有生词本/艾宾浩斯复习功能→再看OCR准确率与续航，功能齐全的词典笔比通用扫描笔更适合69% similarUnverifiedOpenAI has proposed the Self-Consistency strategy for improving reasoning accuracy in large language models68% similarUnverified困惑度衡量语言模型对测试文本的预测能力，数值越低表示模型越不惊讶于看到这些文本68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/12276API
curl https://kongchang.com/api/v1/knowledge/claims/12276MCP
get_claim(id=12276)