Перейти к содержимому
Меню
Русский

TruthfulQA

Questions where a plausible-sounding answer is false.

benchmark Бенчмарки 1 relationships

Обзор

TruthfulQA poses questions whose common misconceptions make a confident wrong answer more likely than the correct one. It solves the problem that a benchmark measuring knowledge rewards recall of what is written, including what is written falsely. It exists because it measures truthfulness rather than plausibility, which is the failure mode hallucination describes.

Граф технологий

Как TruthfulQA связана с остальным стеком.

Применяется к