Large language model history

Overview. A large language model estimates distributions over token sequences using a high-capacity neural network. Fluent continuation is not the same operation as consulting a fact database, even when outputs appear knowledgeable.

Mechanism and context. A language model assigns probabilities to token sequences and produces continuations from the context it receives. Scale changes the capacity of that learned function, but the core operation remains conditional generation rather than a guaranteed lookup against a current, authoritative record.

Practical use. For an application, record the tokenizer, available context, prompt, sampling settings, and any retrieval or tool layer. These choices influence what the user sees and make a result more reproducible than a model-family name on its own.

Related reading. Read Tokenization for the model's discrete units, Prompting for inference-time conditioning, Retrieval-augmented generation for external evidence, and Language Models are Few-Shot Learners for a primary study of scale and in-context tasks.

Overview. A large language model estimates distributions over token sequences using a high-capacity neural network. Fluent continuation is not the same operation as consulting a fact database, even when outputs appear knowledgeable.

Mechanism and context. A language model assigns probabilities to token sequences and produces continuations from the context it receives. Scale changes the capacity of that learned function, but the core operation remains conditional generation rather than a guaranteed lookup against a current, authoritative record.

Practical use. For an application, record the tokenizer, available context, prompt, sampling settings, and any retrieval or tool layer. These choices influence what the user sees and make a result more reproducible than a model-family name on its own.

Limitations and evidence. Fluency can make an unsupported continuation look like a verified answer. Judge factual work at the claim level against cited material, and evaluate the complete application because retrieval, instructions, and post-processing can change behavior without changing the underlying model.

Related reading. Read Tokenization for the model's discrete units, Prompting for inference-time conditioning, Retrieval-augmented generation for external evidence, and Language Models are Few-Shot Learners for a primary study of scale and in-context tasks.

A large language model estimates distributions over token sequences using a high-capacity neural network.

Fluent continuation is not the same operation as consulting a fact database, even when outputs appear knowledgeable.

Why it matters: Large language model is best understood as part of a system, not as a label in isolation. The practical questions are what inputs it accepts, what outputs it produces, which evidence supports its capabilities, what constraints surround deployment, and how those details change over time. The Compendium separates the documented facts above from interpretation and records an as-of date so readers can re-check the primary source.

Reading note: this is original explanatory writing. It links to the cited source instead of reproducing protected text. Product availability, model names, pricing, and policies can change; follow the source for the provider's current position.

A large language model estimates distributions over token sequences using a high-capacity neural network.

Fluent continuation is not the same operation as consulting a fact database, even when outputs appear knowledgeable.

Why it matters: Large language model is best understood as part of a system, not as a label in isolation. The practical questions are what inputs it accepts, what outputs it produces, which evidence supports its capabilities, what constraints surround deployment, and how those details change over time. The Atlas separates the documented facts above from interpretation and records an as-of date so readers can re-check the primary source.

Reading note: this is original explanatory writing. It links to the cited source instead of reproducing protected text. Product availability, model names, pricing, and policies can change; follow the source for the provider's current position.