On "Jev introduces a new shape of LLM - System One, aka Decision Models"

Decision models solving deterministic tasks instead of chasing general capability is exactly the right bet for enterprise search—you don't need a conversational superintelligence to rank results or extract structured data from a page, you need something fast and reliable that does one thing perfectly. The actual opportunity isn't the model architecture itself (that's table stakes), it's whether these things can be materialized into products that beat the current incumbent approach of throwing a big general model at every problem and praying. If this category actually ships as narrow, predictable, cheap inference that outperforms GPT-4 at document classification or entity extraction, SEO tooling and enterprise search systems get better and cheaper overnight—worth watching less for the model itself and more for whether anyone can actually productize the constraint.