实测 astra 干大活:不解释改动理由它就消极怠工
Liu_eroteme · x · 2026-09-08
作者连续几天用 astra 处理大型任务,感受是「又佩服又好笑」:像跟一个心怀不满的资深工程师合作——如果你不解释为什么需要改,它就不太在乎把事情做对。","titleen":"Astra on large tasks: behaves like a disgruntled senior engineer unless you explain why","summaryen":"After using astra on a larger task over several days, the author is both impressed and amused: the model works like a disgruntled senior engineer — if you don't explain why a change is needed, it doesn't really care about getting it right.","channelprimary":"models","channelsecondary":"","tags":["评测","编码","行为观察"],"vendor":"","notificationlabels":[],"xworthy":5,"xangle":"","craftworthy":0,"craftnote":""}
所属事件:GPT-6 Astra 发布：演示惊艳与实测拉胯的落差引争议(10 条相关)→
「模型」频道最新
- php2shell 漏洞实测：仅 2/5 前沿模型能完成利用移植，耗时 3.5-6 小时 — xeophon · 2026-10-27
- Qwen Code 发布 v0.26.2 预览版：新增会话权限规则与实验性共享 Agent Loop — qwen-code-review-bot · 2026-10-27
- GPT-6.1 Terra 事实错误率从 22.3% 降至 4.3%，降幅近 81% — haider1 · 2026-10-27
- GPT-6.1 Terra 发布:早期用户称写作与 computer use 表现亮眼 — BenBajarin · 2026-10-27
- GPT-6.1 Terra 跑分曝光:成本低至 Sol 一半,追不平 Astra — Angaisb_ · 2026-10-27
- Artificial Analysis 更新智能指数 v4.3.2，GPT-6.1 Terra 参与横评 — ArtificialAnlys · 2026-10-27