On LLMs learning new languages

I look forward to a time when large language models are truly “language models”. I think that it is one thing to pre-train a model, and another to actually imbue actual learning and not just memorising at a collosal scale.

I am especially passionate about the ability to actively learn new languages from the user’s end. It could be a minor native language somewhere or it could be a fictional language that the user dreams of.

I am no expert on the matter but this could also apply to creating new computer programming languages to solve a unique problem.