Get the LLM to read and summarise N pages at a time, and store the outputs. Then, you concatenate those outputs into one "super summary" and use _that_ as context.
Theres some fidelity loss but it works for text, because there's quite often so much redundancy.
However, I'm not sure this technique could work on code.
Theres some fidelity loss but it works for text, because there's quite often so much redundancy.
However, I'm not sure this technique could work on code.