WHAT NEW IN X11.05 THU

X 上的 AI 合集档案 · 23 个合集 · 69,408 条

jevPOST

用 Jev 门控抓出 agent 谎报测试通过

agent 声称测试通过却从未调用测试工具,把这次运行交给 Jev 门控读完整历史而非摘要,抓出无依据的结论并转交人工。

NeilXbt
@neil_xbt2026.09.26收录原推发布于 2026.09.25
NeilXbt @neil_xbt原推发布于 2026.09.25
在 X 上打开原推

> Be an agent that says the tests pass. > Never call the test tool once. > Hand the run to a Jev gate. > Let it read the history, not the summary. > Watch it catch the unsupported claim. > Route the run to a human. So why is your agent still grading its own work?

这条是引用着另一条写的:x.com/neil_xbt/status/2102773308672348333

x.com/neil_xbt/status/2103397593607008362采集于 2026.09.26 10:13