Post #2102924

Well, they are language models after all. They have data on language, not real life. When you go beyond language as a training data, you can expect better results. In the meantime, these kinds of problems aren’t going anywhere.