What is an LLM (large language model)?
An LLM (Large Language Model) is a large neural network pretrained on vast amounts of text. It produces sentences by probabilistically predicting the next word (token).
A single model can handle many language tasks, including translation, summarization, question answering, and writing code.
How does an LLM generate text?
An LLM breaks text into small units called tokens, then adds, one at a time, the token most likely to come next given the context so far. It acquires this ability during pretraining on massive amounts of data.
In other words, it does not look up a fixed answer. It generates plausible next words based on probability. That is why it can sometimes state something false with confidence, a problem known as "hallucination."
What kinds of LLMs are there?
They can be divided into open-weight models, whose weights are published, and closed models available only through an API. They are also distinguished by parameter count and by architecture such as MoE.
By use, they include chat models, code-focused models, and small on-device models. A bigger model is not always the best choice for every task.
How do you compare LLM performance and rankings?
Model rankings keep changing with the timing and the evaluation criteria. Instead of memorizing a fixed order, check public benchmarks and leaderboards such as user-vote-based arenas directly.
When comparing, look not only at benchmark scores but also at response speed, cost, context length, and quality on your real tasks.