True but practically training a gpt model is not computationally cheap. I think instead of making such generalized language models we need to focus more one subject specific language models.