推荐janalsncm · 2025-09
“Perhaps this OpenAI paper would be interesting then (published September 4th):”
推荐这篇 OpenAI 论文,讲模型不确定时会猜测
幻觉源于训练与评测奖励猜测、不奖励承认不确定
论文认为幻觉并不神秘:在预训练中它相当于二分类错误,若无法区分错误陈述与事实,统计压力就会自然产生幻觉。后训练阶段幻觉之所以持续,是因为主流评测按对错打分,奖励猜测、惩罚表达不确定。作者主张修改那些主导排行榜、却与目标错位的基准的打分方式,而不是再新增幻觉评测。
评论者引用它解释模型为何宁可猜也不拒答(基准奖励猜测);也有人认为标题比论文结论更耸动,并质疑其中的类比。
“Perhaps this OpenAI paper would be interesting then (published September 4th):”
“This is a de facto false equivalence for two reasons.”
“In part because model performance is benchmarked using tests that favor giving partly correct answers as opposed to refusing to answer.”
“According to that OpenAI paper, models hallucinate in part because they are optimized on benchmarks that involve guessing.”
“It's not as sensational as the headline suggests.”