To my mind, LLMs/genAI is simply irredeemably ugly. It is pure brute force. There is no theory or formal framework behind it: throw enough training data at a model and you can force it to learn anything, in as far as learning means forcing the direction of guessing.
The proponents will point at all the cleverness that has gone in to the model architecture and statistical techniques, but then that goes for a nuclear weapon as well. It's still ugly.
