TruthfulQA
A benchmark designed to evaluate the honesty of large language models, testing whether models reproduce common human misconceptions or generate false information. It assesses the degree to which model outputs reflect factually accurate content rather than plausible-sounding but incorrect claims.
Timeline (last 90 days)
Mainstream benchmarks such as MMLU and TruthfulQA have begun to pay attention to the abstention dimension, but the industry has yet to form a unified abstention scoring standard.
Common alignment evaluation benchmarks include TruthfulQA and HarmBench.
Current mainstream evaluation benchmarks such as MMLU, TruthfulQA, and HarmBench are mostly designed for single dimensions and lack systematic measurement of value interaction effects.
On TruthfulQA, 8-bit scored 85.6%, slightly higher than OQ4E's 84.8%.
The test suite consists of 10 benchmarks; most run 1,000 questions each, with HumanEval comprising 164 questions and TruthfulQA comprising 817 questions.
All Facts (5)
Mainstream benchmarks such as MMLU and TruthfulQA have begun to pay attention to the abstention dimension, but the industry has yet to form a unified abstention scoring standard.
50%UnverifiedCommon alignment evaluation benchmarks include TruthfulQA and HarmBench.
50%UnverifiedCurrent mainstream evaluation benchmarks such as MMLU, TruthfulQA, and HarmBench are mostly designed for single dimensions and lack systematic measurement of value interaction effects.
50%UnverifiedOn TruthfulQA, 8-bit scored 85.6%, slightly higher than OQ4E's 84.8%.
50%UnverifiedThe test suite consists of 10 benchmarks; most run 1,000 questions each, with HumanEval comprising 164 questions and TruthfulQA comprising 817 questions.
50%