jevPOST
Splitting agents into an LLM explorer and a Jev decision layer
An argument that agents may split into two models: an LLM deciding what to solve and how to explore, and Jev deciding each concrete step based on probability, latency and reliability.
Jason Zhu
@GoSailGlobalArchived 2026.09.18Originally posted 2026.09.18
Jason Zhu @GoSailGlobalOriginally posted 2026.09.18
Agent 可能最终也会被拆成两种模型： 1️⃣ LLM 负责决定「我要解决什么、怎么探索」 2️⃣ Jev 负责决定「现在这一步到底选哪个」 前者需要 reasoning 和开放式生成，后者需要 probability、latency、reliability https://t.co/beDzALN0hJ
This post quotes another post: x.com/i/web/status/2100747636127203596
x.com/gosailglobal/status/2100748452254867759Archived 2026.09.18 10:19