Post #3051850

If the creators of AlphaZero had paused the training iterations at each 100 ELO jump, they too could have hyped the fuck out of their latest "o1" reasoning engine and hyperbolized about terraforming space. But no, it is a targeted algorithm based on reinforcement learning (a really goddamn impressive one), but it does not generalize to all of the things. Transformer based neural networks do an incredible job of consuming and producing text, but they are statistical models. It's not magic; it's not even learning or intelligence in the way we use those terms about animals. To pretend that they're going to achieve superintelligence imminently is fuckin ridiculous. Smoke and mirrors from shareholders and industry schmucks There's no reason to think that the timeline for true artificial general intelligence has accelerated so significantly just cause some very smart people have created LLMs that can produce some passable computer code or college essays.