AI Safety Regulations: Prudent or Paranoid? with a16z's Martin Casado

**Erik Torenberg** (0:00)
Hey everyone, Erik here.
At Turpentine, we're building the first media outlet for tech people by tech people. We produce the show you're listening to right now. We're also building a network product that is bringing together the best founders and executives in the game from all stages.
The Turpentine Network is a high trust space where exceptional people can talk. We already have over 400 members and growing. The community benefits include tactical advice, tech stack recs, hiring referrals, in-person events, an investor database, and exclusive perks. If you want to join, you can apply at the link in the description.

**Nathan Labenz** (0:35)
Hello and welcome to The Cognitive Revolution, where we interview visionary researchers, entrepreneurs, and builders working on the frontier of artificial intelligence. Each week, we'll explore their revolutionary ideas, and together we'll build a picture of how AI technology will transform work, life, and society in the coming years. I'm Nathan Labenz, joined by my co-host, Erik Torenberg. Hello, and welcome back to The Cognitive Revolution. Today, I'm excited to share my conversation with Martin Casado, general partner at Andreessen Horowitz. Instead of our usual expert interview format, today Martin and I engage in a genuine back and forth conversation about how we understand today's frontier AI models, the potential for next generation systems to become powerful actors in the world, and what, if anything, we should do about that now.
I had initially planned to discuss our respective AI worldviews only briefly before moving on to questions of appropriate commitments, standards and regulations for AI developers. But we ended up spending a full hour on epistemics and expectations when it became clear that our different outlooks were upstream of any such policy debates.
When we did get there in the second half, I found the discussion far more agreeable than you might expect from today's online discourse. In part, I think, because we had already identified how we understand the situation differently. If I had to try to characterize our perspectives, I'd say that I am more focused on what I can infer from my understanding of current AI systems capabilities and their mechanisms. Well, Martin, who to be clear has definitely also studied today's systems on a technical level, seems to base his expectations more on historical examples and the profound complexity of the broader world. One key question that emerged is how much the potentially irreducible computational complexity of the universe ultimately matters, and whether AI models will be able to take very effective shortcuts relative to full-on simulation.
For several reasons, it seems to me that they probably will be able to do this. For one thing, we humans seem to do it, so there's some existence proof that it's possible. And for another, we see some aspects of these dramatic shortcuts emerging in the models that we've covered in recent episodes on protein folding and interaction, as well as small-scale chemistry simulations. That said, it is still the case that no very general system has developed a reliable world model, to say nothing of an adversarially robust world model at this point in time. And I certainly can't prove that it will happen at any point in time or scale.
While we didn't reach full agreement on worldview, I think this discussion demonstrates that it is absolutely possible to communicate across quite different AI perspectives. And it also reinforces the idea for me that it continues to make more sense to focus our time and energy on understanding what is before jumping to what ought to be done about it. Toward the end, we agreed on the idea of making one or more one-dollar bets about how the near-term future will shape up. I invite any interested listeners to try to formalize those conditions. And if you want to do that on a prediction market platform, I will happily share those links. As always, if you're finding value in the show, we'd appreciate it if you'd share it with a friend. And of course, we welcome your feedback via our website, cognitiverevolution.ai, where we're accepting resumes that I'll be circulating privately with friendly companies. And of course, you're always welcome to DM me on any social network anytime. Now, I hope you enjoy this exploration of AI expectations and implications with Martin Casado, of Andreessen Horowitz.
Martin Casado, general partner at A16Z. Welcome to The Cognitive Revolution.

**Martin Casado** (3:58)
Very happy to be here.

**Nathan Labenz** (4:00)
I think this is going to be a great conversation. We are primarily here to talk about the importance of open source and to dig into a lot of the details of the current debates around early regulation of AI and how that might impact open source and what the pros, if any, and cons, if potentially many of that might be. Before we get into it, I just wanted to get a little bit of your super big picture AI worldview and ask you to project out a little bit.