Context window

The key to getting good results is to provide the LLM with EXACTLY the context it needs. If you want help with a single page of a 25-page document, uploading the entire document reduces effectiveness — it dilutes the content you're targeting.

Break complex problems into intermediate steps with defined checkpoints. To review 10 documents and produce an output, first have the LLM summarize each individually, then feed those summaries into a fresh conversation.

Monitor your context window if the tool allows. When you're approaching the limit, hand off to a fresh conversation with a prompt like: "Please create an output document capturing the key information and insights we have uncovered during the course of this conversation."