Unverified50% confidenceFactExact time
MT-Bench is a multi-turn dialogue evaluation benchmark proposed by Zheng et al. in 2023, containing 80 multi-turn questions and 160 single-turn questions, covering eight categories including writing, role-playing, reasoning, and coding.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEach topic in the roadmap answers four questions: why to learn it, where it is actually used, common beginner mistakes, and a reading guide, with each stage concluding with the construction of a real project.58% similarUnverifiedLLM-as-a-Judge范式由斯坦福大学的MT-Bench与Chatbot Arena研究正式引入学术视野（Zheng et al., 2023），研究发现强大LLM在评估开放式问答质量方面与人类评分者一致性可达80%以上55% similarUnverifiedThe MATH dataset was proposed by Hendrycks et al. in 2021 and contains 12,500 competition-level mathematics problems spanning seven sub-domains51% similarUnverifiedGrice proposed that the cooperative principle of conversation comprises four maxims: quality, quantity, relation, and manner51% similarUnverifiedSelf-consistency sampling was proposed by Wang et al. in 2022: after sampling multiple CoT paths for the same question, the majority answer is taken, which is essentially approximating ensemble learning through sampling diversity.50% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1066483API
curl https://kongchang.com/api/v1/knowledge/claims/1066483MCP
get_claim(id=1066483)