I think the applicability of ML, at least so far, is quite narrow. That's not to say that there are no applications; but rather, that it's not the universal AI

I think the applicability of ML, at least so far, is quite narrow. That's not to say that there are no applications; but rather, that it's not the universal AI tool that it is being portrayed as in the media. ML in practice seems to resemble a complex DSP-like step more than anything. It seems to be mostly used in classical DSP-like domains too (text, speech, images, etc.) It's a tool to handle complex, high-dimensional multi-media data. Though, models like GPT-3 and CLIP show some promise in being something beyond that. With the caveat of having billions of parameters...