# Vals AI estimates Anthropic reaches full recursive self-improvement in 2027

> Original title: 根据 Vals AI 的最新估算，Anthropic 大概会在 2027 年 7-8 月达到完全的递归自我改进（RSI）！Rayan Krishnan 说他们 RSI Index 正在收集一组代理任务，这些任务代表了AI 实验室研究人员在预训练、后训练和框架级工程方面所做的...

- Source: indigo
- Published: 2026-10-28T05:43:14.000Z
- Bad Influence: https://badinfluence.news/items/pxbo7uda4d7tiubis4a8opa38
- Original: https://x.com/indigox/status/2115318454593155077

## Summary

Vals AI estimates Anthropic could reach full recursive self-improvement (RSI) around July or August 2027, according to Rayan Krishnan's account of the RSI Index. The index tracks agent tasks covering pretraining, post-training, and framework-level engineering work done by AI lab researchers. The steepest curve belongs to Gemini, which the author links to Google prioritizing RSI.

## Post

根据 Vals AI 的最新估算，Anthropic 大概会在 2027 年 7-8 月达到完全的递归自我改进（RSI）！Rayan Krishnan 说他们 RSI Index 正在收集一组代理任务，这些任务代表了AI 实验室研究人员在预训练、后训练和框架级工程方面所做的工作。不过最陡峭的增长曲线来自 Gemini，Google 那边应该发生了一些变化，他们开始优先推进 RSI 📈 其实我很看好 Google 的预训练和他们新的组织变化，合并和精简团队、聚焦交付、拿回 AI 前沿，让我们拭目以待👀

[Video](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Famplify_video_thumb%2F2115318400755085312%2Fimg%2F2tAnC04Ct8QkRfAU.jpg&mode=thumb&exp=1793404800&sig=828ed585dc83fdcd)