"Most modern transformer-based LLMs use the same broad structure: tokenization, embeddings, positional encoding, stacked transformer layers (each with multi-head attention and a feed-forward network), residual streams, layer normalization, and next-token prediction."
Similar posts
2 posts close in meaning, closest first
I'm a naturally curious person, so I like to explore any kind of topic. This thread here gives a precise explanation about how LLMs work. In a nutshell, confirms what has been said already: it's controlled randomization of preexisting data. So... math, not chemistry.
The irony with stuff like this is that LLMs are actually unparalleled for *catching* these sorts of errors (if you know how to use them)
