Model Training
Knowledge learned before deployment
- The model learns patterns from large datasets during development
- General knowledge up to a training cutoff — no access to your private data
- Can't be updated in real time without retraining or fine-tuning
RAG
Retrieval Augmented Generation
- Searches external sources (docs, databases) when you ask a question
- Retrieves relevant snippets and includes them in the prompt
- Grounds answers in real data instead of the model's memory alone
Context
What's in the prompt right now
- Everything the model sees for this request — instructions, history, attachments
- Can include pasted text, tool results, or retrieved documents
- Limited by a context window — only so much fits at once