创建

登录 ReadmeX

或

了解我们如何处理你的信息: 隐私政策

ReadmeX

陪你读懂 AI 动态

今天,想先了解什么?

选一份动态总结,或直接聊聊你的问题。

社区和关注动态总结只属于你自己。

资讯

Reddit用户称Astra 6频繁道歉并停滞

AI 总结

一名Reddit用户称,无论上下文长度、任务还是推理温度如何,Astra 6的表现都更像是较旧模型,落后于Claude Opus 5.5和Fable。该用户还报告模型反复道歉、未排查阻塞问题,并回答“实际阻塞仍然存在。我还没有修复它”。

深度见解

这则帖子可能提示Astra 6出现了质量退化,但仅是未经核实的个人使用体验,没有受控测试或开发方确认。

The hell happened to Astra? Astra 6 no matter the context size, task or reasoning temp is a completely different model, its closer to a model form a year ago then anywhere close to the current top models like Opus 5.5 or even Fable.

ive switched to Claude in October when Opus 5.5 was released and was very happy with it, and it has been working on my most difficult ML Projects succesfully, but since i hit my weekly limits often, i thought why not jsut get ChatGPT Plus to finish stuff, they are mostly already 90% finished anyway. Astra couldnt even get started, its one huge apologizing loop, where it messes up, i point out the mistakes and correct, it apologizes, it sometimes apologizes even if it hasnt done anything yet. then jsut stops instead of trying to solve the problem, or ask me for what it needs (EXAMPLE: needs me to log in into a platform, jsut an example, it would jsut stop, not do anything, if then asked why he didnt just ask me to log in he'd yet again, go down a. apologizing rabbit hole) ''you are right, you are right, you are right'' I GET IT ALREADY it literally just repeats the same basic damn AI reply like a model from 2021

#Astra 6#Claude Opus 5.5#Fable#ChatGPT Plus

原文约 2 分钟读完阅读原文
群聊

还没有评论,来说说你的看法。