A large language model is a neural network trained on vast amounts of text to predict the next token in a sequence. That simple objective yields systems that can answer questions, write code, and follow instructions — with uneven reliability and no built-in understanding of truth.
An LLM learns statistical patterns in language by predicting the next token — a piece of a word — billions of times across training data.
Scale matters: more parameters and more diverse data generally improve fluency, but they do not guarantee factual accuracy or safe behavior.
When Latent covers model launches, we separate marketing names from the underlying architecture, context length, and licensing — the details that change who can ship what.
ChatGPT is a product built on LLMs (plus tools, policies, and interfaces). The underlying model family is the LLM; the chat app is the wrapper customers touch.
Typically in stages: pretraining on broad text, then instruction tuning and preference optimization so the model follows useful formats and safety policies.