Large Language Models

Large language models (LLMs) are neural language models trained at scale to predict and generate text, enabling assistants, search, and document workflows.

They are big text-prediction engines that can draft, summarize, and answer questions when prompted.

LLMs are Transformer-based (or related) neural language models trained primarily via next-token prediction at scale, then often instruction-tuned and preference-optimized. Capabilities depend on data, alignment, tooling, and evaluation—not model name alone.

LLMs power assistants, document workflows, and developer tools. Separating the model layer from application wrappers improves vendor and architecture comparisons.

• Enterprise assistants and copilots
• Document Q&A and summarization
• Code assistance under human review

• Flexible natural-language interfaces
• Rapid prototyping of language workflows
• Strong few-shot adaptation via prompting

• Can hallucinate unsupported facts
• Context window and cost constraints
• Require grounding, evals, and access control in enterprises

• Fluency is not the same as truth
• An LLM is not automatically an agent with tools
• Open weights do not remove licensing or safety duties

What is the difference between Generative AI and LLMs?

Generative AI includes many modalities. LLMs are language-centric models that are a major—but not only—generative class.

• NIST AI RMF

• Vaswani et al. — Attention Is All You Need (Transformer)