The efficiency of these systems is improving at an incredible speed. You can run local Qwen today on a consumer grade laptop that beats a frontier model from less than a year ago. In a year or two, local models will be the norm for programming. The whole hyperscale compute thing is just a brief mainframe era of this tech. So, using that as an argument is just silly.

Meanwhile, all content is human made because LLMs are just a tool that a human uses. This whole narrative is frankly bizarre beyond belief.