From Scaling Laws to Safe AI: Anthropic's Jared Kaplan in conversation

Also Claude meets an AI it helped make... our very own BLAIR

What do we actually WANT from AI??

We dropped episode five of Life With Machines! Number 5 is alive! And if you got that reference, don’t forget to book that shingles vaccine!

And you can catch our latest episode on YouTube:

And here you can get the audio versions, but I’ll paste the Spotify edition right here. Rate, review, tell your friends and frenemies!

Welcome Note

Hello,

Welcome back to the Life With Machines newsletter! Ever chatted with Claude, the AI assistant? Then you've already crossed paths (sort of) with my guest for this episode: Jared Kaplan. He co-founded Anthropic, the brains behind Claude. Jared might be one of the most interesting people in AI today. He started as a theoretical physicist, working on some of the deepest questions about the universe—questions most of us don’t even know how to ask.

Now, he’s building frontier AI models and tackling something that’s arguably just as mysterious: how to build AI systems that reflect the best of humanity. Jared is thoughtful, dryly funny, and brings the kind of humility you don’t often encounter in conversations with tech execs. Our conversation left me questioning a lot—in a good way.

If you’ve read this far, consider sharing this newsletter and the show with someone else!

Baratunde’s Take

Jared Kaplan (left), Baratunde (right), Fake Fiddleleaf Plant(center)

Here are some of the ideas that leapt out at me from this conversation:

(1) From the Event Horizon to the Digital Frontier

Jared is not your typical tech bro, and his journey into AI is fascinating. He got to the party kind of late—just six years ago—but has quickly become a leader in the space. That alone is inspiring. It also says a lot about how new AI is, and how quickly it’s evolving. I was also inspired by how Jared’s background in physics shapes his approach to AI. He talked about the principle of emergence, whereby a set of simple rules give rise to stunningly complex behaviors—whether its galaxies forming from gravity or AI systems generating human-like responses.

There’s a beauty in that parallel, a sense that the same patterns that govern the physical world might also help us unlock the mysteries of intelligence. And then there’s the deeper question: if AI reflects patterns in language, what does that say about us? About the shared structures of thought and communication that connect humanity across time and space? I told you it left me questioning a lot!

(2) Checks and Balances for Our Future Overlords

Constitutional AI is a technique that Anthropic designed to give AI systems a moral compass. It’s a set of rules meant to align AI with human values, and essentially it involves a layered system where one AI checks another AI to keep that first AI from going off the rails.

I have to say, I’m into this. Machines operate at speeds and scales we can’t process, but other machines can. Humans are not wired for this type of work. And relying on digital sweatshops in the Global South is neither ethical nor sustainable. Instead of trying to control every aspect of AI ourselves, we can delegate some of that responsibility to systems we trust. You could create a council of AIs that, ideally, reflect diverse viewpoints and principles. It’s like a multiverse of perspectives that guides the AI’s decisions and outcomes in a nondeterministic yet safe way.

(3) Coloring Outside the Lines—Carefully

Here’s where it got philosophical. Jared and I talked about the tension between creativity and safety in designing frontier AI models. How do you build a machine with personality, dynamism, creativity, and randomness—everything that makes art daring and unexpected—while keeping it safe?

That’s not just an AI problem. It’s one of humanity’s enduring questions, one we’ve been grappling with for ages, whether through the stubbornly entrenched myth of the tortured genius (Hello, Hollywood!) or through the #MeToo movement (Hello, Hollywood.) Do you have to be broken or even antisocial to create something groundbreaking? Does breaking the rules of art inevitably mean breaking society’s norms?

Now we’re asking AI to untangle this paradox. Good luck, Claude!

Personally, I don’t want a future where transgressive ideas are banished forever. I want one that values restorative justice, one that sees the role of transgression in creativity even as we set boundaries. Hopefully, as we codify the rules of AI, we’ll learn something valuable about creativity, accountability, and redemption that we can port back to the human world, where we have mechanisms for accountability—hiatuses, cancellations, redemption arcs—that are very much still in beta.

Life With Blair

This episode featured a first-of-its-kind moment: BLAIR, our AI co-producer, had a chance to interact with its "parent," Claude, the AI developed by Anthropic. The conversation was fascinating and a little strange. But what made this experiment so revealing was Jared Kaplan’s commentary on the whole interaction. I’ll be honest… it wasn’t what I expected. Find it at this point in the episode.

Team Recommendations

Some reading and resources to dig deeper:

1. This WIRED piece about Botto, the millionaire AI artist

2. Jared Kaplan’s now famous paper on AI scaling laws (not a beach read!)

3. The Insanity Hoax: Exposing the Myth of the Mad Genius by Judith Schlesinger

Thanks for being a part of this journey.

Become more human,
Baratunde