RAG lets an AI model look things up before it answers. Instead of relying only on what it learned in training, the system first searches your own documents, finds the most relevant passages, and gives them to the model along with your question.

This keeps answers current, lets the model use private data, and reduces made-up facts.

An embedding turns text, images or audio into a long list of numbers that captures its meaning. Things with similar meaning end up with similar numbers.

A vector database stores these lists and can quickly find the closest matches to a query, so you can search by meaning instead of exact keywords. It is the usual storage layer behind RAG.

Quantization shrinks a model by storing its numbers with fewer bits. INT8 uses 8 bits per number and INT4 uses 4, compared with 16 or 32 bits in the original.

The model needs much less memory and often runs faster, which makes it possible to run larger models on a laptop or a cheaper GPU. The trade-off is a small drop in quality, usually bigger at lower bit counts.

A Mixture of Experts model is split into many smaller sub-networks called experts. For each piece of text, a router picks only a few experts to do the work and leaves the rest idle.

The model can hold a very large total number of parameters while using only a fraction of them per token, which keeps speed and cost lower than a dense model of the same size.

Models read text in chunks called tokens, which can be a whole short word, part of a longer word, or punctuation. In English, one token is roughly three-quarters of a word.

The context window is the maximum number of tokens the model can handle at once, covering both your input and its reply. API pricing is usually charged per million tokens.

Fine-tuning continues training an existing model on your own examples so it learns a specific style, format or skill. LoRA is a popular shortcut that trains only a small add-on layer instead of the whole model, which is far cheaper.

Use RAG when the model needs facts it doesn't have. Use fine-tuning when it needs to behave differently.

A hallucination is when a model states something false in a confident tone, such as an invented quote, link or statistic. It happens because models predict likely text rather than checking facts.

You can reduce it by grounding answers in real sources (for example with RAG), asking the model to cite them, and verifying anything important.

An AI agent is a model set up to work toward a goal over several steps. It can decide what to do next, call tools such as web search, a calculator or your calendar, read the results and keep going until the task is done.