Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Jeff Dean and Fei-Fei Li weigh in on AI self-improvement

AI summary

At Stanford HAI’s “Confronting Our AI Future” conference, Jeff Dean and Fei-Fei Li discussed the boundary between ordinary self-improvement and recursive self-improvement (RSI). Dean said automation can shorten experiment iterations, but training, computation, data handling, and validation impose physical limits, making unlimited acceleration unrealistic. The conversation also covered Discovery Loop, where Dean said roughly 20 automated improvement loops are running to explore faster experimentation and learning from earlier results.

In-depth insight

The discussion frames RSI as an engineering problem with concrete constraints rather than an unlimited self-accelerating loop, while providing current detail about Discovery Loop’s automated research approach.

在斯坦福HAI的闭幕主题演讲中,Jeff Dean介绍了他离开Google后参与创立的新公司Discovery Loop。它的使命是构建能自动解决机器学习、科学和工程重要问题的AI系统:提出实验、运行实验、评估结果、决定下一步,并让每一轮实验反哺下一轮。随后,李飞飞与Jeff Dean围绕“自我改进”与“递归自我改进”的区别、AI解数学难题带来的冲击、Transformer的起点、技术预测的对错,以及这一代年轻人面对AI时的焦虑展开对话。

以下是他们的对话实录:

李飞飞:先谈谈Discovery Loop。为什么现在要成立这样一家公司?

Jeff Dean:我们的使命是构建AI系统,自动解决机器学习、科学和工程中的重要问题。我们相信,这最终会为世界带来更多发现。很多学科都有一个共同的循环:提出实验,实施并运行实验,观察结果和测量,再决定下一步,通常就是做另一个实验。我们想自动化这类实验循环。如果能缩短迭代时间,自动提出并运行大量实验,从每次评估中学习,让后续实验更好,同时提升实验质量和数量,就可能得到突破性结果。

#Jeff Dean#Fei-Fei Li#Discovery Loop#Stanford HAI#Google#Gemini#Anthropic#METR

Original · 11 min readRead the original
Group chat

No comments yet. Start the conversation.