| To do it right we have to dispense with old ideas about why we’re doing it to whittle down what to preserve. We don’t need to preserve every cat picture and vacation moment, just enough geometric and color data to replicate them. Same with text. Deduplicate to an essential data model of knowledge and algorithms to boot strap that into a useful state. Stable Diffusion runs in a browser on a phone. I have no doubt we can get a “bigger model” out of a tiny model and a few algorithms that will hydrate it and update its state based upon a context. But IT is too literal to serve “business” so we end up with the janky mess you called out. So in summary; a rethink of our economic approach to keeping people busy, and compressed, deduplicated model that fits on a phone, with hardware and software that know to get what the user asks for out of this small hash. That’s what all the ML is leading to. It’s just like I said; IT is a boring industry to serve business traditions. Intentionally: https://www.escapistmagazine.com/Bobby-Kotick-Wants-to-Take-... I’m not that smart and figured this out. Asked a colleague with a PhD in math working at MS who was all wink and nudge at me. The “smarter people” know what’s up and want to be the first one there. Then under politically correct traditions it’s “their property.” It’s not that hard to conceive; all information displayed on a machine is 2D geometry and games simulate 3D through well understood perspective and color theory. We just need a model that captures all the coordinates of our text and those perspective shifts and color compositions. But to get there you need social change to avoid the loop you say you’re tired of but no one else does. Good luck. |