A large language model is trained to predict text, and from that objective it acquires broad capability to summarize, answer questions, extract information, and draft. It processes input within a bounded context window and produces output token by token.
Its knowledge comes from training data with a cutoff date, which is why systems handling current or matter-specific facts supply documents at question time rather than relying on the model's memory.
Alternative Names:
LLM
Why it Matters?
The critical property for legal work is that these models generate plausible text rather than retrieve verified facts. That is the mechanism behind fabricated citations, and it is not a defect to be patched but a characteristic of how the technology works. Architectures that supply source documents and require citation to them are the practical response, along with verification before reliance.
Frequently Confused with
Related terms
Frequently asked questions
Why do language models produce false citations?
Does a bigger model eliminate errors?
