Dwarkesh Podcast Progress Update

"I was 2 months away from quitting the podcast for 2 years."

Can we just take a step back and acknowledge how crazy it is that this is my job?

I get to read whatever interests me. Then I get to ask every question I’ve accumulated to the world's smartest person on that topic. Not only that, but there are hundreds of thousands of people out there who appreciate the output of this process and get some kind of value out of it. 

In hindsight, I keep forgetting how many times I almost quit the podcast to start (without loss of generality) a GPT wrapper. I'm reminded of this every time I talk to a friend who knew me a year ago, and they say, "Good to see the podcast is doing well. I remember you telling me you were gonna quit that and do something else. Still planning on doing that?" 

I had put off applying to jobs before college ended, so when I hit graduation, I was totally unsure of what to do next. Around this time, Anil Varanasi, who was apparently one of the few hundred listeners of my show at the time, asked me how much it would cost for me to do the podcast/blogging stuff full time for 6 months. I said $10,000 (which in my mind was a highball since I was living with my parents in Texas and had negligible expenses)¹.

I want to emphasize that the podcast was not on people's radar as something super promising. It wasn't even on my radar that way. I was just trying to find my next thing. About a year ago, Anil told me (somewhat) jokingly that one of his biggest regrets in life was giving me this grant, since it resulted in me wasting my talents on some tiny podcast instead of pivoting away to a more ambitious project. He’s changed his mind since, and I appreciate that he’s one of the few people I can count on for honest, unvarnished advice. But this candid take even from my earliest backer shows you how fledgling the whole project was up until very recently.

A couple months later, I wrote a blog post that got Jeff Bezos to say nice things about me. So then I put off quitting the podcast for a couple months. Even then, for at least another year, I was definitely planning on winding down the podcast and starting a startup. I moved to SF and started hacking together some basic GPT wrapper demos. I would tell myself, “I have a podcast interview scheduled with someone I’m really excited about, but after that I’m gonna quit.”

I was 2 months away from quitting the podcast for 2 years.

Frequency and flying monkeys

I’ve been thinking about how many interviews I should be doing. Right now, I average one every other week. That’s a slow pace given that this is my full time job. Why am I not doing an episode week - or even multiple episodes a week? Certainly not because I’d run out of interesting people to talk to. The amount of people out there who know things I want to learn more about might as well be infinite.

If I did more episodes, I could be more risk taking with the kinds of people I talk to. The obscure scholar who might either become the next Sally Paine, or wilt under the camera - the homeless guy who might have fascinating stories to tell, or may just go schizophrenic - the amateur travel writer who has seen the world but might not be able to engage in conversational repartee. Right now, if an episode like this turns out to be a dud that I don’t air, it represents a couple weeks worth of wasted preparation, travel, production, etc. If I was recording 1-2 episodes a week (with minimal overhead on me personally), this wouldn’t be such an issue. I could take so many more shots on goal.

And there’s another reason why I want to move through my queue of potential guests faster. More than once, some legendary octogenarian whose work I’m a huge fan of, and who I had vague plans to interview “someday”, died before I sent them the cold email. Most recently, this happened with James Scott, author of Seeing Like a State and Against the Grain - he passed away a month ago. Some of the greatest scholars alive are quite old, and this adds a lot of urgency. I don’t want to miss the opportunity to get as much out of these legendary minds as I can, while I can. (TODO: Gotta get to Robert Caro somehow!)

So what’s the constraint on me doing more episodes? The obvious answer is ‘quality’. Could I host great interviews, and do the prep which I’m known for, if I couldn’t sink 2 weeks into every guest? 

If I’m being honest with myself, I think I could. A lot of time goes into the “everything else” bucket - selling recording ads, micromanaging clips and edits, logistics for setting up shoots, following up with people, and all other kinds of flying monkeys. I have a long todo list this week, and less than half actually corresponds to some kind of reading or writing.

Suppose I was able to offload all of this. I come into the office in the morning, I read and write all day. The next day, I step into the studio room to record an interview, and then immediately get back to reading for the next guest. If everything else is taken care of, and I’m being honest with myself, I could record two episodes a week. 

Now this doesn’t give me any time to write, which I want to make a big priority over the next year. But that takes us down to one episode a week, which is still twice as frequent as my current pace. 

My current producer (as well as my editor) is truly amazing. I would be producing an episode a month without him, seriously. But his summer break is ending in a month (he’s an incoming sophomore). I need to hire someone else by then who can buck shot all these fucking flying monkeys.

Monetization

Ads are becoming quite lucrative. I only take ads from companies that I think are great, and I say no to most potential sponsors. The problem is that this kind of due diligence takes a lot of time. And I don’t want to spend a couple days a week moonlighting as an ad salesman and company investigator. These advertisers are paying me a lot of money, and in return I think it’s really important to do everything - to make sure that the way hundreds of thousands of people learn about their company and brand is one they’re extremely happy with. 

But the end result is that once you include the time to negotiate and sell an ad, write a bespoke script for the episode, and then refine the copy, and then recording the ad, and then re-doing the script based on feedback and re-recording, and working with my editor to add graphic overlays, I'm looking at close to a day+ of work per ad.

So I've been asking myself, well, if ads are going to take this much time anyways, wouldn't it be better to use that time to make content for paid subscribers? In fact the very reason these advertisers want to sponsor my podcast is because I have a great audience with lots of potential users. But what product could possibly have better audience fit on my podcast than paid subscriptions for my Substack? And if this framing is correct, then my audience is suffering a big deadweight loss in the form of the time I could be spending making paywalled content, but which I'm instead spending selling and recording ads.

Even if this is much less profitable to begin with, it’s a valuable hedge. And who knows? Ads ended up being more lucrative than I anticipated - maybe the same could be true of paid subscriptions.

Choosing guests

My job is shockingly open ended. Next week, if I wanted to, I could interview a historian of antiquity about whether disease caused the fall of the Roman Empire, a primatologist about what changed with human brains, or a battery engineer about the pace of electrification. 

A lot of great people ask to come on the podcast. But I usually say no. I choose guests not based on whether I want to spend two hours chatting with them, but whether I would learn a lot by spending two weeks preparing. It’s a high bar, and it’s often uncorrelated with how famous they are. Based on the analytics, I can tell you the audience cares very little. My most popular guest is the historian Sarah Paine - she was (in the best sense of the term) an obscure intellectual and a hidden gem. The episodes I’ve done with my friends Trenton, Sholto, and Leopold have beat out some CEOs of famous companies. 

Jeff Bezos has a famous quote:

When somebody congratulates Amazon on a good quarter, I say thank you. But what I'm thinking to myself is— those quarterly results were actually pretty much fully baked about 3 years ago. Today I'm working on a quarter that is going to happen in 3 years.

I feel like a similar thing is true about the podcast. A great episode is baked in thanks to the reading I was doing 6 months ago. Right now I’m digging the rabbit holes which will pay out dividends during the interview I record 6 months from now. It’s always very valuable to keep a bunch of background processes running. I had been reading about World War II and military history for months before I was even thinking about doing an episode on those topics. At some point, I asked on Twitter who I should interview about WW2, and Tanner Greer gave Sarah Paine a rave review. She’s now my most popular guest.

Learning

There are many podcasters who have been interviewing experts in every domain for years and years. And they don’t seem that much smarter as a result². You can’t just hope to become some giga-generalist by chatting up a bunch of experts. The default path for a podcaster is that you don’t learn that much over the years and spend most of your time on topics which are easy to talk about in a shallow way (corruption of institutions, culture war, etc).

I want to do a different thing. I want to actually improve over time. I want to develop my own takes. I want to understand everything. And this is how the podcast improves over the long run too - by me becoming smarter and learning more shit. That is the underlying beta. Everything else is a temporary benefit at best (and more often, a total distraction).

How do I learn more things? I should write more. Many of the most important questions simply can't be addressed extemporaneously over a podcast.

Besides, I don’t think I can improve as an interviewer if I’m not publishing ideas of my own. The AI interviews I've done after writing my scaling post have been far better than the ones I did before writing it. I sometimes wonder if my current method of preparing by reading piles of only potentially relevant material in the hopes of conjuring up random questions once in a while is not that productive. Maybe I should instead just spend 2 weeks writing my take on the most interesting question in the guest’s field. And ask them, “Now explain to me why I’m a total moron - let’s hash this out.”

What topics in particular to keep learning about in the background? I found that some topics - history especially - are very amenable to casual reading. Whenever you're bored, you can just pull out a book about the history of oil or the Bronze Age Collapse or something.

But many important subjects (particularly those on the frontier) don't have this quality. You're not just going to accidentally stumble upon a deep analysis of solar deployment, synthetic fuels, or DNA foundational models. If you want to understand these topics, you have to make it your job for a couple of days to weeks. I’m lacking in these kinds of topics. Which is a real shame, because these frontier technologies are the most important things happening in the world - you can’t develop a coherent worldview and predictions about the future without understanding them.

You can’t even fully understand AI without understanding these other topics (for example: Can the 1GW cluster be powered by a 20,000 acre solar farm? What kinds of physical manufacturing could a GPT-5 level AI do? How baloney is the AI biorisk misuse stuff?)

Another category of topics which I’ve neglected that are especially relevant to thinking about AI: Is a nationalized AGI project a good idea? Who exactly would run it? And: What’s actually going on in China? A lot of people talk about it as if it’s some ant-colony-like totalitarian blob which could pour arbitrary trillions of $ at ASI whenever they want. But is that how China actually works? I have no idea what’s actually happening there, and  I don’t think that the AI researchers (in whose worldview a China race is so load-bearing) do either³.

Anyways, this is all to say that I’m going to travel to China in November to find out more for myself.

Other thoughts

• Wasted time is measured not in afternoons spent dithering around, skimming around different books on Kindle, and taking leisurely walks. Wasted time is measured in weeks spent shoveling bullshit, answering emails, micromanaging the insignificant. Wasted time is measured in the months spent working on an ill considered and inferior project. 

• I usually feel disappointed by how little I got done in a given day or week. But I feel really proud of what I achieved over, say, the last year. I'd much rather have it this way than the opposite.

• Over the last couple months, I’ve gotten 3 friends to be tutors for me. Basil Halperin is my econ tutor, Kevin Kiningham is my hardware tutor, and Adam Scherlis is my physics tutor. I’m now more convinced than ever that Bloom 2 Sigma is real. And the curriculum they compile is as valuable as the session itself - as a self-directed learner, it’s super useful to have a subject level expert tell you what the key papers/textbook chapters/etc are in the topics I’m most interested in.

• The reception to my episode with Trenton and Sholto was pleasantly surprising to me. After we finished recording, I told them that this was my favorite episode so far, but that in all honesty, it might not end up being that popular because it might have been too technical for most people. Instead, the episode went viral. It’s become Dwarkesh Podcast canon, and it’s more popular than the 2 episodes I released before that one with famous CEOs. Part of the explanation for why that episode did so well is just that we are good friends and the vibes were awesome. But part of it is that people want to be treated to the conversation that I would actually have with them at dinner. Intuitively, it feels like I should do the,”Wait, can you explain this for the audience first before we go deep into it?” But I get the sense that immersion learning is super effective and way more fun, and the “Let's backup and explain for the audience” stick is annoying and patronizing. 

1. A month later, Effective Ideas offered me another small grant, which gave me some more runway. Thank you to Anil and Effective Ideas!

2. The exception that proves the rule is Tyler Cowen. He hosts the best interview podcast out there because most of his job is reading, writing, coming up with takes.

3. Replace China with America in this sentence, and you realize how naive this picture looks. “America is gonna build a trillion dollar cluster!” Okay but who exactly? Biden? Microsoft? And with whose money? I get that they have a different system, but they’re still individual human beings with their own local incentives, factions, beliefs, constraints, etc. They’re not a billion piece uniform Lego set which Xi Jinping can turn into a well oiled ASI factory.