Large language model
Also: LLM
A very large neural network trained to predict the next token of text — the thing meant by 'AI' in most 2020s headlines.
Last updated
A large language model is a neural network trained on huge corpora of text to predict the next token — a sub-word fragment. Because that one training objective generalises, a sufficiently large model can answer questions, translate, summarise, classify, and draft prose. The “large” is literal: modern LLMs have tens of billions to trillions of learned weights.
Why it matters for your research. Almost every sentence about “AI changing humanities research” is, under the hood, a sentence about LLMs. Understanding that an LLM’s output is a fluent statistical guess — not a lookup against sources — is the foundation for judging claims about provenance, citation, and reproducibility.
In Archēglyph. We do not use LLMs to answer questions over your corpus. Where we invoke an LLM-adjacent tool — for example a vision-language model on a page image — the output is a structured decision (region kind, OCR text), and the exact model id is recorded. See Transparency is a feature.
Not to be confused with. Model is the broader term — an embedding model, a classifier, and an LLM are all models. LLM is one specific kind.