WHAT NEW IN X11.02 MON

AI collections from X · 22 collections · 65,263 posts

jevIMAGE

JevHarness lets LLMs generate task-specific Jev decision flows

JevHarness lets an LLM generate a task-specific decision flow once and freeze it, raising win rate on a Pokémon task from 25% to 75% over five iterations.

SuSu_酥酥👅
@NFT_ChenArchived 2026.09.22Originally posted 2026.09.21
JevHarness lets LLMs generate task-specific Jev decision flows (image 1 of 3)
JevHarness lets LLMs generate task-specific Jev decision flows (image 2 of 3)
JevHarness lets LLMs generate task-specific Jev decision flows (image 3 of 3)
SuSu_酥酥👅 @NFT_ChenOriginally posted 2026.09.21
Open original on X

🔥牛逼!让 Jev 自我优化、自己解决复杂问题,不用再手写 instruction、criteria、候选集! JevHarness:LLM 自动生成任务专属决策流程,冻结后每次只跑代码 + 快速 Jev 判断。有数据还能用轨迹持续进化。宝可梦 5 轮迭代,胜率 25% → 75%。 主要亮点: 🔹LLM 写一次策略,Jev 做高速模糊判断
🔹自动生成特征、问题、判据和动作逻辑
🔹可用奖励 + 全轨迹反射 / GEPA 自我优化
🔹自带 Coding Agent skill,丢仓库地址就能开干 🔹可回放每一步 Jev 选择的对战 Demo 关键操作 1️⃣仓库:https://github.com/TianyuCodings/JevHarness 2️⃣Claude Code 装 skill:
/plugin marketplace add https://github.com/TianyuCodings/JevHarness.git
/plugin install jev-harness@jevharness 3️⃣对 Agent 说:用 JevHarness skill,按我的任务生成 harness 4️⃣有奖励就开评测 + 迭代;满意后冻结,线上只跑 harness + Jev 5️⃣Demo:https://jev-harness.tianyuchen99.chatgpt.site/ #Jev #JevHarness #AgentHarness #开源 #LLM #决策模型 #宝可梦AI #自我优化

This post quotes another post: “Jev decision model on OpenRouter: 154ms median latency and 98.5% accuracy”

x.com/nft_chen/status/2102005334827204978Archived 2026.09.22 10:16