Raw LLM Responses

I'm aware of all of that. But this post is addressing the most cutting edge AI platform at the time of writing this which cannot scale as you mention, AKA: chatGPT. And will probably still apply once GPT-5 comes out and so on. I'm aware of locally sourced AI models, but I addressed those points on the "Some other company will just make "unlocked ChatGPT"" section of the post (see "use a more limited or cheap AI model") I'm sure this will change over time as you say once AI models become better and more power and resource efficient, but that's a separate topic for the future.