No. Don't do this again.
I'm not sure you understand what you're saying here. Actually, I am sure that you don't.
Sophy is the end-product of a machine-learning process. It learns how to get around a race track from cues (pointing the right way, staying on the track, not hitting things) in a reward-loop. Every successful loop is "rewarded" and incorporated into the next iteration -while unsuccessful ones are discarded - until it reaches the desired level of competence. That set of instructions is then locked in to what's called an "agent", and that's what you race against. It requires some compute power, but no more than running a few hundred PS5s for a while, and the complete set of agents requires one PS5 to run each time.
Generative AI is the end-product of applying autocorrect to copyright theft. It scans existing creative works - writing, art, music, video - to build a library of patterns in order to guess what comes next; the more advanced it gets, the better it gets at guessing. Every step of the process requires an incredible amount of compute power (hence massive data centre farms, buying up every bit of silicon it can and even those that haven't been made yet), and the final/current version of it runs in live. Creating a single GenAI image from a prompt takes up approximately 500ml of water (in cooling at the data centre) and enough electricity to propel an electric car a few hundred feet; text is less, video is much, much more.
They're both "AI" but they're not anywhere near the same thing, and only one is relevant here.
