Post #2345347

@TheEvilSkeleton@social.treehouse.systems Well, I've seen an IRC chatbot built of simple Markov chain models. You feed it a corpus of text — you can fully control what you put in it —, and it learns, effectively, to do word-by-word autocomplete on that corpus, with some optional randomness. If you want to, you can add in some grammring in order to enforce structure. That's nothing like an LLM (at least inside), but it's computationally a lot cheaper, doesn't require clusters unless your corpus is ridiculously large, doesn't require graphics cards.