We've already seen zero-shot learning of novel syntaxes (that is, chatting with an LLM to interactively devise a new language that is then used, without any act

We've already seen zero-shot learning of novel syntaxes (that is, chatting with an LLM to interactively devise a new language that is then used, without any actual retraining of the model).