I think the biggest issue with the AI models is that we/they call them AI models. They are just using a different way of processing the data. This isn't an endorsement or indictment of these models, but, I am a firm believer that hardware and software engineering has progressed enough since the first numerical models were developed that augmenting them with training data and neural processing, things that weren't available a decade ago, seems like a worthwhile endeavor.
We had AI before chatbots became mainstream and it was called "machine learning", and the newer models are much closer to that than having a conversation with ChatGPT and it suggesting you kill yourself because you may not get 6" of snow again until the year 2028.