WHAT NEW IN X11.06 FRI

X 上的 AI 合集档案 · 23 个合集 · 69,408 条

gemini-4IMAGE

Gemini 4 Argon 真实代码库修 bug 实测失败,Opus 5.5 成功

开发者给 Argon 一个真实工作 bug 并指出相关文件,它仍翻查 30 多个无关文件,还编造出不存在的数据库索引列;同一任务 Opus 5.5 修好了。

Lakshya
@Its_lakshya_ai17 天前收录原推发布于 2026.10.19
Gemini 4 Argon 真实代码库修 bug 实测失败,Opus 5.5 成功
Lakshya @Its_lakshya_ai原推发布于 2026.10.19
在 X 上打开原推

GEMINI 4 ARGON JUST FAILED THE TEST THAT ACTUALLY MATTERS !!!!! Not the benchmark Not the leaderboard Real-world coding... A developer gave Argon a genuine work bug and even pointed it toward the files where the issue was likely hiding Argon still went through 30+ irrelevant files 😭 Then it got worse… It confidently reported a database index column that literally didn’t exist 💀 Same task with Opus 5.5? Fixed. And this is exactly why I’m skeptical of benchmark hype A model can top every leaderboard But if it confidently invents things inside a real codebase, developers stop trusting it FAST Argon might be insanely powerful But power without reliability is a dangerous combination Maybe the real benchmark is simple: Can I trust you with my codebase? 👀

x.com/its_lakshya_ai/status/2111883201031606585采集于 2026.10.19 09:58