Wouldn't it become more and more difficult to improve the specialist model? And wouldn't the ultimate generalized models (AGI) eventually be able to improve the

Wouldn't it become more and more difficult to improve the specialist model? And wouldn't the ultimate generalized models (AGI) eventually be able to improve themselves automatically? Or am I talking about something different here?