Glossary
Terms, in one line each
Quick lookups. Linked terms have a full knowledge base page.
- Agent
- A model running in a loop: think, call a tool, look at the result, repeat until the task is done. Covered in episode 9.
- Context window
- Everything the model can see for a single request: instructions, conversation, documents, tool results.
- Embedding
- A list of numbers representing the meaning of a piece of text, used to find similar content. Covered in episode 6.
- Eval
- A repeatable test of AI output quality, made of test cases plus pass/fail criteria. Covered in episode 12.
- Fine-tuning
- Extra training on your own data to specialize a model. Usually not the first tool to reach for.
- Hallucination
- When a model states something false or made-up with confidence. Covered in episode 2.
- Inference
- Running a trained model on a request. The model does not learn during inference.
- Knowledge cutoff
- The date the model's training data ends. Newer information must come in through context.
- LLM
- Large language model: a system trained to predict the next token of text, which in practice lets it write, summarize, reason and code.
- MCP
- Model Context Protocol: an open standard for connecting models to tools and data sources. Covered in episode 10.
- Next-token prediction
- How an LLM generates: predict one token, append it, repeat.
- Non-determinism
- The same input can produce different outputs, so you design for a distribution of results.
- Parameters
- The billions of internal numbers a model adjusts during training. Often called weights.
- Prompt injection
- Malicious instructions hidden in content the model reads, trying to hijack its behavior. Covered in episode 20.
- RAG
- Retrieval-augmented generation: fetching relevant documents and putting them into the context before the model answers. Covered in episode 6.
- Stateless
- The model keeps no memory between calls. The full conversation is resent each time.
- Streaming
- Showing tokens as they are generated instead of waiting for the full response.
- System prompt
- Instructions the product places at the start of the context to set the model's role, rules and style. Covered in episode 4.
- Temperature
- A setting that controls how much randomness goes into choosing each token. Low is predictable; high is varied.
- Token
- The chunk of text a model reads and writes, roughly ¾ of an English word. Cost, speed and limits are measured in tokens.
- Tool use
- Letting the model call functions or APIs you define, so it can take actions and fetch data. Covered in episode 7.
- Training
- The phase where a model learns by adjusting its parameters on huge amounts of text. Done by the AI lab.