3 ms·
If the context window is finite, then LLMs actually are Markov chains. It's just that they're a much more efficient way of representing transition probabilities
by c1ccccc1 4y ago
If the context window is finite, then LLMs actually are Markov chains. It's just that they're a much more efficient way of representing transition probabilities than storing them all in a giant lookup table.
- nextaccountic 4y agoWhat's a LLM? It isn't a logic learning machine, right? Or a load linked Markov You mean, a large language model?