This paper tests whether AI can realistically guess what a specific social media user would comment when they see a new post.
Large reasoning models can often find the right math answer in their โheadโ before finishing their written steps, but this works best in languages with lots of training data like English and Chinese.