Original Summary

Posting here as I haven&#x27;t found the answer anywhere.<p>I know that GPT Astra uses looping instead of reasoning tokens. My specific question is this: reasoning tokens pollute the context and makes compaction kick in quickly. Does this also happen to looped transformers like Astra?

  • 情报分类：技术学习与提效
  • 分类依据：内容涉及技术、AI、软件工具或工程实践
  • 信息来源：Hacker News 新项目
  • 发布时间：2026/10/4 03:32:58