Skip to main content
Capability Doctrine

What is a large language model?

A large language model is an AI trained on vast text to predict the next word in a sequence. That makes it fluent and general, but it generates plausible text rather than looking up facts, which is why it can be confidently wrong.

5 min read/Written by Perry Luzier/Reviewed

Prediction, not retrieval

An LLM answers by generating the most likely next token given the input, repeated until it has a full response. There is no lookup step, so accuracy depends on whether the facts were in training, or supplied at query time.

Because the model produces text statistically, its confidence is unrelated to its correctness, it sounds equally sure whether it is right or inventing. This is the core insight operators miss: the fluency is not evidence of accuracy. Reliable systems therefore do not trust the model to know; they supply the facts it needs at the moment of the question, which is exactly what retrieval-augmented generation does.

Questions

Frequently asked questions.

Are bigger LLMs always more accurate?

Not for your specific facts. A larger model is more fluent and general, but it still will not know your internal data. Grounding a mid-sized model in your documents usually beats a giant model guessing from memory.

Want this built into your operation?

We install the systems described here as owned infrastructure. Start with a diagnostic of where your business actually loses time and margin.