Large Language Model (LLM)

Level 1

A super-smart computer that has read almost every book and website, so it can talk and write like a person.

Level 2

A really big AI trained on tons of text from the internet that can write essays, answer questions, and have conversations.

Level 3

A neural network with billions of parameters trained on massive text corpora to understand and generate human language. Examples include GPT-4, Claude, and Gemini.

Level 4

A transformer-based model trained on web-scale text data using self-supervised learning, capable of in-context learning, instruction following, and multi-step reasoning across diverse language tasks.

Level 5

An autoregressive or masked language model leveraging the transformer architecture at scale (10⁹–10¹² parameters), exhibiting emergent capabilities like chain-of-thought reasoning, tool use, and cross-domain transfer as a function of compute and data scaling laws.

Sources

• NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.