ChatGPT picks differently when asked in a different language. "Choose a random fruit from this list" asked a LOT of times
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
OpenAI 相关模型动态已经出现，适合跟踪能力变化、生态影响和后续可用性。
一项针对 OpenAI GPT-5.6-Luna 的实验显示，当以不同语言提问时，模型选择水果的方式有所不同。用户用英语、波兰语和日语，共 6000 次要求模型从给定列表中随机选择一种水果。尽管提示语和水果顺序相同，但模型始终只选择了四种水果。随后，用户将水果列表打乱，再次运行了 6000 次实验，结果略有改善，但仍仅选择了三种水果。…
I did a little experiment and asked OpenAI's GPT-5.6-Luna 6000 times to randomly pick a fruit from this list:
And across the three languages I picked (English, Polish, Japanese) only four fruits were ever picked. The wording of the prompts and the order of the fruits were the same across all three languages. I reran this another 6000 times with the fruit list shuffled each time, and you can see the results of that on the second chart. Again only three fruits, but the results are a little better now.
(Pardon the lychee and pomegranate icons, there's no emojis for these two fruits, and of course the model picked these, lol)