Reference · 15 terms
Glossary
Every short form used across the series. Inside a lesson, hover or tap any dotted term to see the same entry in place.
LLM features
- circuit breakerCircuit breaker
- After repeated failures, stop calling a dependency for a while and fail fast.Lesson 4 →
- excessive agencyExcessive agency
- Giving an LLM more tools, permissions or freedom than the task needs.Lesson 0 →
- golden setGolden set
- A fixed set of example inputs with known good answers, used to test an LLM feature.Lesson 2 →
- improper output handlingImproper output handling
- Trusting model output without checking it before passing it to a browser, shell or database.Lesson 0 →
- kill switchKill switch
- A runtime setting that turns a feature off in seconds without a deploy.Lesson 4 →
- least privilegeLeast privilege
- Give a component only the access it needs for its job, and nothing more.Lesson 1 →
- LLM-as-judgeLLM as judge
- Using a model to grade another model's answers. It is useful but has biases, so check it against human grades.Lesson 2 →
- model routingModel routing
- Sending easy requests to a cheap model and escalating hard ones to a stronger model.Lesson 3 →
- prompt cachingPrompt (prefix) caching
- Reusing the work done for an unchanged start of a prompt, so repeat requests are cheaper and faster.Lesson 3 →
- prompt injectionPrompt injection
- Text in the input (or in content the model reads) that tries to change the model's instructions.Lesson 1 →
- redactionRedaction
- Masking private data such as emails and card numbers before storing or sending text.Lesson 4 →
- regression gateRegression gate
- A check in CI that fails the build if quality drops below a set level.Lesson 2 →
- spanTrace and span
- A trace is the record of one request. A span is one timed, labelled step inside it.Lesson 4 →
- tokenToken
- A small piece of text a model reads or writes. Cost and delay follow token counts.Lesson 3 →
- TTFTTime to first token
- How long until the first piece of an answer arrives.Lesson 3 →