Post #1613214

>Then along comes the language model. Suddenly, you just talk to the computer the way you'd talk to another human, and you get what you ask for. That's not at all how LLMs work, and that's why people are saying this whole premise is a bad take. Not only do LLMs get things wrong, they do it in such a way that it completely fabricates answers at times; they do this, because they're pattern generation engines, not database parsers. Algorithms don't do that, because they digest a set of information and return a subset of that information. Also, so what if algorithms cost a lot of money? That's not really an argument for why LLMs level the playing field. They're not analogous to each other, and the LLMs being foisted on the unassuming public by the billionaires are certainly not some kind of power leveler. Furthermore, it takes a fuckton more processing resources to run an LLM than it does an algorithm, and I'm just talking about cycles. If we went beyond just cycles, the relative power needed to solve the same problem using an LLM versus an algorithm is not even close. There's an entire branch of mathematics dedicated to algorithm analysis and optimization, but you'll find no such thing for LLMs, because they're not remotely the same. No, all we have are fancy chatbots at the end of the day that hallucinate basic facts, not especially different from the annoying Virtual Assistants of a few years ago.