Jack Rogers

If you actually believe that major AI dev enterprises are afraid of their own toys, and demanding that the US government regulate them, then please dial the number listed below. ... And we will connect you to a representative from the Federal Witless Protection Program. This is Silicon Valley we are talking about, the reigning all-time champions of hubris and lack of perspective. Any one of these guys would burn the world if it meant he could be king of the ashes that remained. No, they want a regulatory moat. A bunch of federal bureaucracy and compliance requirements that they can cut through with money, which will hopelessly entangle the open-source models following close behind them. Why? And more importantly, why now? Because AI is not on the verge of superintelligence at all. Quite the opposite. It's hit a point of diminishing returns. LLMs are great with language now, but they lack theory of mind, executive function, and a world-object model, and those are not little glue-on doodads. They are problems as deep as language processing, which humanity hasn't even scratched the surface of yet. Several more fundamental paradigm-shifting breakthroughs will have to occur before humans even build sapience, much less superintelligence. And AI companies cannot pull these innovations out of their collective asses, because it is the very nature of innovation to show up unexpectedly and by accident. So why is this a crisis for OpenAI, et al, and why does it put them in danger from open source source models? It's because of what I call Second Implementer Effect. It's closely related to Second System Effect. Ever notice how the first-to-market company doesn't always win in the long run? Ever notice how a second company frequently comes along and eats their lunch? To develop something entirely new requires a lot of resources. A lot of investment. A lot of engineers. And that means that once you actually build it, you are a big company with a lot of engineers. And they need something to do. You can't just trim sails and fire them, because you have a lot of investors and need a lot of revenue to pay them. Line must go up. So you build lots and lots of new features, for which you can charge more money. You become Oracle. But 99% of the market doesn't want Oracle. Customers don't want infinite features for infinite money. They only want more up to a certain point. After that, they want those features, and only those features, for as little money as possible. And what often happens is that your first implementer shoots right past that point, gold-plating the product until nobody wants to pay for that much gold. Then the second company, which just offers a basic product at a basic price, eats their lunch. Because the second implementer is the right size for the market. In this case, the risk to OpenAI, et al, is that open source models will be that second implementer. Nobody is gonna pay giant data centers by the GPU cycle if they can have "good enough" running on their home machine. And if AI have hit diminishing returns, or soon will (yes), then they can't widen or maintain the gap by speeding up. They have to slow everyone else down. You can bet they are burning up the phone lines to their pet senators. That's why everyone is suddenly telling you the singularity is tomorrow. Precisely because it's not. The Navier-Stokes thing is a double nothing burger, two juicy helpings of nothing between two slices of nothing, with no cheese, no lettuce, and no tomato. Have you noticed that no one is telling you about all the wonderful new technologies that will result from the corner case they brute-forced? Because there aren't any. Its a mathematician's toy, and if you want to know more about how and why, read what Feynman wrote about Banach–Tarski, which will give you the general idea. Does this mean AI is useless? Far from it. Large Language Models have already proven useful, for things like research, personalized instruction, coding and so forth. And we've only begun to scratch the surface of what we can do by communicating commands to a computer in natural language. Cool things will continue to come out of this. But they will happen at the interface and application layers. No superintelligence will be involved. For the immediate future, AI will remain the paintbrush, not the artist. And, yes, there are also risks. AI can pretend to be human, and persuade retards of all sorts of things. But that's a human failing. It's not the AI's fault we let the retards vote. That's how we got a whole bunch of midwit senators who want to "regulate" innovation, instead of medicare fraud, invading third world scum, and invading third world scum who commit medicare fraud.