But maybe the dataset for said language is just too small to produce correct content at all, and that's why ChatGPT is bad, not because "ChatGPT is thinking in

But maybe the dataset for said language is just too small to produce correct content at all, and that's why ChatGPT is bad, not because "ChatGPT is thinking in English". Except when it default to answering in English, ChatGPT sounds like french when I talk to it in french. The bigger the dataset, the best is it's capabilities, training it exclusively with a language don't mean it will be better at this language.