jevPOST
实测：Jev 做浏览器操作和 LLM 评审都行
作者跑完真实评测，认为 Jev 在 browser/computer use 上表现好，替代 LLM-as-a-Judge 更强，下一步做 LLM 干活、Jev 决策的混合 agent。
Michael
@MichaelAzmy2026.09.23收录原推发布于 2026.09.22
Michael @MichaelAzmy原推发布于 2026.09.22
Ran real evals: Jev is great for browser/computer use and even better as an LLM-as-a-Judge replacement. Next experiment: a hybrid agentic harness where the LLM does the work and Jev handles decisions, classification, and routing.
x.com/michaelazmy/status/2102354813807964485采集于 2026.09.23 10:17