I haven’t watched it yet but given the context I’m going to assume they mean ‘traditional’ machine learning and not specifically the LLMs or diffusion models which are now usually called “AI”. I feel like domain specific ML models are not a problem - they don’t scrape the web or require new power plants to operate.

But yeah LLMs are evil.

And I’ll see if I’m wrong after watching it.

Update:

Yeah, I cringed a bit every time he said AI for a regular machine learning system, but that’s what he meant. And unfortunately given the general public I kinda understand why he used the term. (And he does actually explain it as ML also)