Transcription: OAPN Episode 226 - Worse is Better

Host: Happy Monday, welcome to work. Okay? Get to work. No, here we go. We got--where's--we got some furniture going. And, uh, we've got over here, the world's best and worst production studio. Yes. Are you ready to record yet? Let's do it.

Guest: No.

Host: No. No. I gotta kill some time for a bit. Skip ahead.

Carr: And we're back. Episode seven, lucky number seven.

Host: And our inaugural episode from our new production studio in downtown Austin. Thank you, Carr, for setting this up. It's literally the perfect studio, because it's both the worst production studio in the world and also the best. You'll see.

Carr: For the record, I only had like five hours.

Host: This has come together, it's--no, it's amazing, it's amazing. Um, but guess what? Like, it's part of the worst AI lab, called OpenAgents, but which is also the best AI lab. And you'll see why shortly. Um, okay, so this article we're gonna go through a little bit and--and see why I'm saying this. Um, you'll forgive us for our inaugural episode here, getting a little philosophical about what we're actually doing here. And, uh, this will tie into Bell Labs in a cool way so you can kind of see why we're making some of the choices that we are. Um, so this article was kicking around the early Nostr scene, the Telegram groups 2021, 2022, uh, to sort of explain the unique characteristics of Nostr and why we thought it was poised to grow. And it's kind of, um, some excerpts from, like, I don't know, chronicling of computer science history over the last few decades. Um, and this distinction that it introduces between the New Jersey school and the MIT school, I think is really, uh, illustrative of our approach, which is New Jersey versus I think the big labs are sort of this MIT approach. And you're going to increasingly see, um, conflict between OpenAgents and the sort of truly open AI community more broadly. And this is kind of, like, a cool lens to--to interpret through some--so, uh, I'm gonna read through a little bit of this real brace--real briefly. So, the MIT slash Stanford style of design is captured by the phrase "the right thing." You're going for simplicity, correctness, consistency, completeness, but like with an emphasis on, like, designing the right thing and then everything has to use the right thing. And they give as an example these languages: Common Lisp with CLOS and Scheme. That represents the MIT approach to design and implementation. And you may say, "What the heck are those?" Now we're going to compare that in a second to, okay, "The worse-is-better philosophy is only slightly different." Simplicity, correctness, consistency, completeness, but with an emphasis on starting simple and letting it grow. Early UNIX and C, have you heard of those?

Carr: Yeah.

Host: You heard of these? Only, like, the most popular operating system and programming language of all time that are like the bedrock of, like, a--like, global civilization. UNIX and C are the New Jersey approach. And to jump ahead real quick, uh, New Jersey. Why is it called New Jersey? Guess who's in New Jersey. Uh, Bell Labs. Where these things were invented. And this approach came to be called the New Jersey approach. If you want a cooler name for it, sorry New Jersey, call it the Bell Labs approach, okay? So, back to my article here. Uh, so the worse-is-better philosophy, he's kind of being funny here that it's worse, but he believes that even worse is--worse-is-better even in its strawman form has better survival characteristics than "the right thing" and that the New Jersey approach when used for software is a better approach than the MIT approach. He goes into a whole story and if you're, you know, a fellow computer nerd you should read it, but like the headlines here are: UNIX and C are the ultimate computer viruses. Uh, once it gets out and spreads, then it will upgrade to, like, take on whatever characteristics it wants from "the right thing." So, the lesson to be learned from this: it is often undesirable to go for "the right thing" first. It is better to get half of "the right thing" available so that it spreads like a virus. Once people are hooked on it, take the time to improve it to 90% of "the right thing." The parallel here in Nostr where there's been other people, companies who've come along and said, "Nostr sucks because of" fill in the blank, "and my version's going to be better because I figured out this, this, and that that the Nostr people are too dumb to think" and like, multiple people have said that. And the adoption numbers for all of those protocols combined are nowhere close to Nostr. Because Nostr is this New Jersey approach. And like, "Oh, let's do the MIT approach." And we really draw lessons from this. It's like, how can you build a thing that catches on? So, in the case of OpenAgents, like, what are we not doing? We're not spending 12 months, you know, raising money and hiring people and putting them in a room that you never hear from for a year because they're building some secret super intelligence behind closed doors that at some point will be released on the world. That's, like, every other lab is doing some version of that or did versions of that. OpenAI back in the day like, you start by going over--but like, we're not doing that. We're starting with, like, what is the thing that's going to grow? How do we start by optimizing for virality? I know, Carr, you've got some compute there. Can I pay you Bitcoin for it?

Carr: Sure.

Host: How much you want?

Carr: I don't know.

Host: Exactly. You don't know, because no one else is buying it. We're the only buyer. So, if the 20 gig--just imagine--I'm not gonna pull it up again. The graph of the 20 gigawatts of consumer compute relative to the 2 gigawatts of OpenAI's compute, there's no buyers for that massive thing. The price of it is zero. We are putting a price on it. So to begin with, we have this essentially limitless pool of compute accessible to us if we can sustainably purchase it from you. Now, us giving you Bitcoin for a thing, of course you're happy to sell it to us. The big question, and what we've spent a lot of our time on is, can we produce something that's economically valuable that just is better than the cost of essentially nothing. It's like, can we do anything at all that's valuable because if we do, we've got a flywheel that's going to begin. And it's not going to be perfect. It's not going to have all the, like, tens of billions of dollars of budget of the big products, but if we've got a simple basic interface that you're happy to click around and earn some Bitcoin and then be productive with, um, we've got the makings of something truly great.