Can Humans Dream of Electric Sheep? Human-Written Samples for Fine-Grained Vision-and-Language Hallucination Benchmarking

AI 中文解读

核心亮点：人类写的“错误答案”也能用来测试AI的“幻觉”问题了，而且效果不输AI自己生成的，这让AI评测变得更客观、更持久。通俗解读：AI经常“一本正经地胡说八道”，比如看着一张猫的图片，却硬说成狗，这叫“幻觉”。以前测试AI有没有这个毛病，得靠AI自己生成一些胡说八道的例子，但不同AI说的瞎话不一样，标准很难统一。现在研究人员干脆让真人来写这些错误描述——用了1600条真人写的样本，涵盖中英法意四种语言，还对比了五款主流AI生成的18400条样本。结果发现，真人写的错误描述和AI自己犯的错在类型上非常接近，而且真人写的更规范、更容易控制，用来当“考卷”反而更靠谱。实际影响：以后AI公司发布新模型时，我们不用再担心它是不是只会在自家考题上表演“标准答案”了。有了这种真人编写的测评集，AI的“胡说八道”会更容易被发现，你用AI查资料、认图片时，看到的结果会更可信。尤其对需要多语言支持的用户来说，中文、英文、法文、意大利文都能测，覆盖面更广，用起来更放心。

In an age of rapid model turnover, how do we make hallucination evaluation more perennial? We explore whether human-written hallucination samples could take the place of model-generated hallucinations, in order to make benchmarking detection independent of particular models. To this end, we construct a dataset of 1,600 human-written samples, spanning four languages (Chinese, English, French, Italian), and 18,400 samples from five vision-and-language models, all annotated for hallucinations using a fine-grained span-level labeling scheme. We find that human-written samples result in higher agreement and allow greater control of dataset contents, while remaining distributionally similar to samples derived from vision-and-language samples and providing a reasonable portrayal of detection capabilities - suggesting that human data is a viable substitute for model-based hallucination benchmarks.