Large Language Model

Definition

A large language model is a learned system that estimates relationships between tokens and uses those estimates to continue, transform or classify sequences. The word “large” usually refers to some combination of parameter count, training data and computational scale rather than a strict threshold.

Operating model

Text is segmented into tokens, mapped into numerical representations and processed through repeated transformer layers. Attention mechanisms let the model weigh relationships between positions in the available context. During inference, the model produces a probability distribution over possible next tokens and a decoding strategy selects from that distribution.

Scope

The concept includes pretraining, supervised adaptation, preference optimization, prompting, context windows, tool use, retrieval augmentation, quantization and evaluation. It does not imply that every model has the same architecture or that text generation alone establishes factual reliability.

Electronic Artefacts position

Electronic Artefacts treats the LLM as one component inside a wider knowledge or creative system. Sources, retrieval, permissions, tools, interfaces, provenance and human review determine whether the model is useful and governable.

Limitations

Language models can produce plausible but unsupported statements, inherit dataset bias, lose information outside their context, expose sensitive inputs and behave differently under small prompt changes. Their outputs require evaluation appropriate to the domain.

References

See Vaswani et al., Retrieval-Augmented Generation, Generative AI, Provenance and Augmented Intelligence.