LLM
An LLM (large language model) is a model trained on very large volumes of text, able to process and produce natural language. It underlies mainstream generative AI services.
What it does, and does not do
An LLM predicts the continuation of a sequence of text. It does not consult a knowledge base, verifies nothing, and has no notion of accuracy. Plausibility is precisely what it optimises, which is why fluent output can be entirely wrong.
Cost
Training consumes considerable resources, concentrated in time. Inference, the use of the trained model, is far cheaper per request but is repeated at every use. For a service in production, inference eventually exceeds the cost of training.
The size of the model is therefore the main lever. A smaller, specialised model frequently meets the need at a fraction of the cost, both financial and environmental.
