A staged compression idea for very long AI sessions
Very long AI conversations or work sessions can be handled by reading them in stages instead of forcing the whole record into one prompt. The method starts with a heavily version so the model can build a broad outline. It then reads less slices to add more detail.
At the end, it can read small verbatim chunks when exact detail matters. The idea borrows from , where something starts blurry and becomes sharper, but it does not use the formal math behind s. Each slice is so it fits inside the , meaning the model only needs the current slice, the user request, and the current draft output at one time.
The model is also told which pass it is on, so it knows whether to make an outline or fill in details. The main goal is to preserve non-local information, meaning patterns or that only appear when the whole session is considered.
Key points
- Long AI sessions are read in stages with different levels of .
- The model first builds an outline from a version, then adds detail from less slices.
- Each slice is small enough to fit inside the .
- The model is told which pass it is on so it can switch between outlining and adding detail.
- The method tries to protect non-local information that may disappear when a session is split into small pieces.