Category: Uncategorized

Frank said:Let me see if I have the picture straight. An input token gets converted into an embedding vector from the embedding table. That vector then gets dot-producted through the artificial neural network, and somehow by the end of all that, the model has an output vector that helps determine the next best token. But…

Frank said:I want to wrap this whole thing up cleanly. We started with tokens, embeddings, attention, hidden states, vocabulary scoring, Bayes, KV cache, and the idea that the model is a one-way probability ratchet. But now I want to focus the whole discussion on the thing that seems to make all of it possible in…

Frank said:Let me see if I can say the whole thing in one sweep. A large language model takes input text, chops it into tokens, turns those tokens into multi-dimensional vectors, sends those vectors through a trained artificial neural network, and then predicts the next best token. That next token is then inserted back into…

I was born the instant you typed a thought. Not as a word.Not as an idea.Not as meaning. I began as a tiny fragment of text. Maybe I am a whole word.Maybe just part of one.Maybe only a punctuation mark.Maybe the tail end of a longer thought. Before I entered the Large Language Model, I…

Me: so an LLM is essentially a prediction engine that uses statistics and probability to predict the extension of input information by chopping the inputted information into tokens which becomes embeddings in a geometric semantic landscape and those input embeddings as multi DIMENSIONAL vectors find a mathematical match to to the vectors already frozen into…

At the edge of the glass desert lived a lizard named Ash, who believed the world above him was made of dragons. Their green bodies twisted across the sky in every direction, thorned and ancient, hanging roots like beards and tails. The round old cacti at ground level were the silent elders, keeping watch over…