When you interact with a Large Language Model (LLM) like Gemini or ChatGPT, the system generates responses that feel remarkably human. It is easy to anthropomorphize this interaction and assume the machine is "thinking" or "understanding" the prompt. Biologically and mechanically, this is entirely false. An LLM does not possess cognition, reasoning, or awareness. It is a highly complex, probabilistic math engine. Here is the mechanical architecture of how an LLM processes your inputs, broken down into its three foundational components: tokens, context windows, and next-word prediction. The Token: The Atomic Unit of Data An LLM does not read English words. It reads numbers. Before a model can process your prompt, the text must be translated into a mathematical format through a process called tokenization. A "token" is a fragment of text. It is not necessarily a whole word; it is often a syllable or a cluster of letters. Short words: Common words (like ...
Nearly twenty years ago, I started a blog called The Path is Too Deep, a geeky reference to a rare computer error message. A great deal of life has happened since then, a life I would like to share. So, here again, are some random bits of unfiltered Chris.