I Think It's a Good Idea to Apply a Decision Transformer to a Language Problem

It's quite possible that the language model doesn't know enough to solve that problem correctly alone. Ironically, a lot of the more anti deep learning cognitive psychologists in the old days were very focused on language and the idea that everything that humans could do took place in like, a language of thought. And so i guess if you're jerry photer, you shouldn' basically think that a good enough language model, which may be existing ones aren't, together with the right incoters and decotes, is a human level intelligence. Think language is very limiting in some ways. Uns finding the right balance between using language and not forcing the system to use language too strongly is tricky.