Pinned post
Introducing a16z Originals: Working Theory.
I sat down with @johnschulman2 and @markchen90 to demystify RSI: AI helping build better AI.
Models can already run experiments. But research taste—the judgment to know which questions are worth asking—remains harder to teach. As John points out, knowing the literature isn’t the same as having been wrong and learned from it.
We also explored how AI could help solve alignment, and why the benefits of RSI don’t depend on fully autonomous research.
I came away optimistic about AI’s potential to accelerate discovery, including solutions to some of the problems it creates, and believe that human judgment remains essential for the foreseeable future.
Huge thanks to Mark and John for joining us.
I sat down with @johnschulman2 and @markchen90 to demystify RSI: AI helping build better AI.
Models can already run experiments. But research taste—the judgment to know which questions are worth asking—remains harder to teach. As John points out, knowing the literature isn’t the same as having been wrong and learned from it.
We also explored how AI could help solve alignment, and why the benefits of RSI don’t depend on fully autonomous research.
I came away optimistic about AI’s potential to accelerate discovery, including solutions to some of the problems it creates, and believe that human judgment remains essential for the foreseeable future.
Huge thanks to Mark and John for joining us.
For decades, the idea of AI improving itself was just a thought experiment. Today, AI is already helping researchers at frontier labs run experiments, write code and test ideas faster. So how far are we from AI that can do the research on its own?View quoted post on X
In our first episode of Working Theory, a16z General Partner Sarah Wang explores recursive self-improvement with two leading AI researchers: Mark Chen, Chief Research Officer at OpenAI, and John Schulman, co-founder and Chief Scientist of Thinking Machines Lab, and co-founder of OpenAI.
They cover what RSI actually means, why research taste is so hard to teach, and why AI may be one of our best tools for solving alignment.
00:00 The last invention
01:58 What is RSI?
03:21 How far are we?
05:56 Getting to full RSI
10:57 Aligning RSI
15:08 Working theory
YouTube: youtu.be/h_aA1-R70E8
@markchen90 @johnschulman2 @sarahdingwang