I am also trying to understand the difference between compaction, and what IDEs like Cursor do when they "summarize" context over long-running conversations.
Is this saying that said summarization now happens at the model level? Or are there other differences?
Is this saying that said summarization now happens at the model level? Or are there other differences?