LLM
Large language model: an AI model trained on very large amounts of text to predict and generate language, which can be used to answer, summarise, write, reason about and act on text.
A large language model is a neural network trained on very large collections of text to predict what comes next. From that simple objective, scaled up, it learns to answer questions, summarise documents, translate, write code and follow instructions. Models such as the GPT family, Claude and Gemini are LLMs; the broader term foundation model also covers systems trained on images, audio and other data.
What an LLM is, and is not
An LLM on its own produces text from its training and its prompt. Most of what makes an AI product useful and reliable comes from the system built around it: retrieval (RAG), tools, memory, checks and the harness that coordinates them. This is also where the main risks sit, such as hallucination.
Evaluating one
A model name is not a complete description of what you will run. The same label can be served through different routes and wrapped in different harnesses, so evaluate the configured system, not the label: its runtime address. Systems that use LLMs to plan and act over multiple steps are discussed under agentic AI.
Related terms
RAG
Retrieval-augmented generation: a pattern where a model is given relevant documents or data retrieved at question time, so its answers rest on current, specific sources rather than only its training.
Hallucination
When an AI model produces fluent, confident output that is false or unsupported by its sources. What matters is which errors are unacceptable in a given workflow and how they are detected.
Harness
The software around a model that manages context, tool use, sub-agents and the execution environment. It shapes real-world capability, so it must be evaluated with the model, not ignored.
Agentic AI
AI systems that pursue a goal by planning, using tools and taking multi-step actions with some autonomy, rather than only answering a single prompt.
Used in these essays
Stop adopting AI
AI adoption is rising faster than enterprise value because companies keep installing new intelligence inside old operating models.
Cahit Arf’s 1959 question about thinking machines
In a 1959 text on thinking machines, Cahit Arf moved from clocks and relays to a harder question: what happens when a machine meets a problem that was not anticipated when it was built?
You don't know what you don't know. Here's how to start.
The most dangerous moment in an AI transformation isn't when you make a wrong move. It's when you don't know you're making one.