 |
1. Context Window as a desk
An LLM's context window is like a desk; it's the space where the model processes information. The size of the desk determines how much information the LLM can handle
|
2. Prompts and Responses
The prompt you provide the LLM is a piece of paper on the desk. The LLM will consume this information and present its response as other pieces of paper of the desk.
|
 |
|
3. Grounding with RAG
When using a RAG (bookshelf) the LLM has relevant information from the RAG provided via post-it notes within its context (i.e. on the desk). The threads you see from the desk to the bookshelf are links that help it ground to known sources.
|
4. Accessing tools.
When the LLM has acces to tools, the method in which they can be accessed for their functionality is also provoided within the context. Their usage adds to the space consumed on the desk.
|
|