got any .pdf's of the rules? I can PM you step-by-step instructions (or just post here) how to set it up. If you've got an m-series macbook you should be able to run at least a 4b models, and if you have a pro or m4 macbook with more memory you've got a pretty capable machine.
this is all accurate, and I think I would also add to your points: genAI ultimately produces an "average" of everything it's read and ingested. It definitionally is not going to be creative or groundbreaking, but it certainly can do a lot of heavy lifting and sifting of unstructured information.
What I'm interested in technologically is for LLM's to be applied to more than just lingustic information. What happens when you develop a neural network that has a foundational and didactically unique understanding of traffic flow engineering, or chip design, or any variety of fields that involve more data than we raw meatbags are capable of processing at scale.
None of that is big and sexy, but it is extremely powerful when applied to the right problem set. Once we get to a point of being able to train narrow field expert models, and then scalably orchestrate them together into a larger mixture of experts model (similar to the DeepSeek architecture) that would really be one hell of a powerhouse model.
....
It would probably take a dyson sphere level of energy to produce such a generally capable model though. At least at the current (known) levels of computational power and solutions available