Learn the basics of LLMs in 60 seconds with Beau Carnes

[00:00] Let's learn the basics of LLMs in less [00:02] than 60 seconds. LMS are next word [00:05] prediction machines. That's it. They [00:07] guess what comes next. First, they train [00:10] on the entire internet. Books, articles, [00:13] code, conversations, trillions of words. [00:16] Text gets split into tokens. Words or [00:19] pieces of words. The model only sees [00:22] numbers. The transformer architecture [00:25] lets it look at all tokens at once. It [00:27] learns which words relate to each other. [00:30] When you type a prompt, it calculates [00:32] probabilities. What's the most likely [00:35] next word? It picks a word, adds it to [00:38] the context, then predicts again, one [00:40] token at a time until it's done. That's [00:43] it. Training tokens, attention, and [00:45] prediction. It's not thinking. It's [00:48] pattern matching at an insane scale. and [00:51] you just learned the basics of how LLM's