# DeepMind高管谈AI递归自我改进与基建投资

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-08-05 20:35
- AIHOT 分数：42
- AIHOT 链接：https://aihot.virxact.com/items/cmsg2ly2a03drrolgq6zvyw9t
- 原文链接：https://x.com/rohanpaul_ai/status/2084981660924383465

## AI 摘要

Google DeepMind首席战略官Jasjeet Sekhon称，AI基础设施支出正为期望中的自我改进循环提供资金。递归自我改进（RSI）最理想形态下，AI将以更少人类监督设计、评估和训练更强后继者。当前证据仍有限，如Google AlphaEvolve提出的算法改进使Gemini训练时间减少1%，属有界改进而非自主设计后继者。

## 正文

"If you hit recursive self-improvement, that curve will go to hyperexponential, and that is a key part of the investment thesis, the scientific thesis, and a key part of why society's investing what it's currently investing in."

Google DeepMind Chief Strategy Officer Jasjeet Sekhon says AI infrastructure spending is financing a hoped-for self-improvement loop.

In its strongest form, recursive self-improvement (RSI) will let AI design, evaluate, and train more capable successors with progressively less human supervision.

The current evidence is still narrower: e.g. Google's AlphaEvolve proposes algorithms, then automated evaluators run and score them against human-defined objectives.

Google says its resulting kernel changes reduced Gemini training time by 1%, an example of bounded improvement rather than autonomous successor design.

----

From "Berkeley RDI" YouTube channel, (full video link in comment)