Core vocabulary
Words made simple
A few useful meanings before you read the explanation.
- LLM
- A model trained to work with language. It can write or summarize text, but its answer still needs checking against reliable information.
- Token
- A small piece of text processed by a model. It may be a whole word, part of a word, or punctuation.
- Attention
- A model mechanism that combines information from different parts of the input when deciding what to produce next.
More words from this lesson
- Embedding
- A list of numbers that represents the meaning of some content. Similar meanings can produce nearby lists of numbers.
- Vector
- An ordered list of numbers. An embedding is a vector used to represent content for tasks such as similarity search.
- RAG
- Find relevant information first, then give it to a model to help answer the question. The answer is only as reliable as the information and checks used.
- Retrieval
- Finding and fetching information that is useful for the current question.
- Structured output
- An answer organized into specified fields, such as claimId and status, instead of only free-form text.
- Use this after studying the full explanations.
- Spend 10 minutes on vocabulary, 10 on RAG and agents, 10 on security/workflows/evaluation, and 10–30 on the insurance design and scenario answers.
| Concept | Say this in an interview | Remember the limit |
|---|---|---|
| LLM | A trained model that processes/generates language using learned patterns and supplied context | It does not automatically know or verify private account facts |
| Tokens | Model processing units used for context limits, throughput, and often billing | Words/characters only estimate usage |
| Context | Information supplied for the current model call | Large capacity is not perfect attention or permanent memory |
| Prompting | A clear task contract with inputs, constraints, output, and unknown behavior | Prompts do not enforce backend permissions |
| Zero-shot / few-shot | Instructions alone / instructions plus examples | Evaluate example quality and token cost |
| Structured output | Model data constrained to a declared schema where supported | Valid shape is not factual or business validity |
| Tool calling | Model proposes a named operation and arguments; application validates and executes | Never grant direct broad database or network authority |
| Embedding | A vector representing useful content features for similarity | Similar meaning does not prove truth, access, or applicability |
| Cosine | Similarity of vector direction | Score is not probability of correctness |
| Vector DB/index | Storage/indexing for similarity retrieval and metadata | Filtered recall and lifecycle matter |
| BM25 | Lexical relevance ranking over matching terms | Good for identifiers; paraphrases may need semantic retrieval |
| Hybrid search | Combine lexical and vector candidates | Fuse ranks or calibrated scores, not arbitrary score addition |
| Reranking | Carefully order a retrieved candidate set | Cannot recover missing candidates |
Can you explain it without the notes?
Core vocabulary
Use your own words. Explain an example and one limitation before checking the explanation.
Check your explanation
Use this after studying the full explanations. Spend 10 minutes on vocabulary, 10 on RAG and agents, 10 on security/workflows/evaluation, and 10–30 on the insurance design and scenario answers.
Turn this concept into a clear answer.
Practice this conceptAdapted from 13_AI.MD · “Core vocabulary”. Examples are study fragments; framework behavior can depend on your version.