SCOTT DETROW, HOST:
This week, AI giant Anthropic pulled off the largest IPO in history. Just three years ago, the company was in a cramped office with an uncertain future. This week, on Machine Gods, a new podcast from NPR all about the AI revolution, hosts Kevin Roose and Casey Newton look back at the early days of Anthropic and why almost no one thought it would succeed.
(SOUNDBITE OF ARCHIVED NPR CONTENT)
CASEY NEWTON: Casey here with a quick disclosure at the top. As regular Machine Gods listeners know, my fiance works for Anthropic.
(SOUNDBITE OF MUSIC)
NEWTON: So I want to get into this, Kevin, because you were one of the earliest reporters to take this company seriously. You wrote a really long and great profile of them for The New York Times in 2023. And I'm curious what your initial impressions were and whether any of those were notably right or wrong.
KEVIN ROOSE: Yeah, so I went back in my notebook this week to the interviews that I did for that piece in 2023, and I am shocked at how many scoops that I just sat on and did nothing with. I was like...
(LAUGHTER)
ROOSE: I was like, they told me during these interviews that every Anthropic co-founder had the same amount of equity. I completely forgot that, and...
NEWTON: (Laughter).
ROOSE: ...Reported it as if it were new information when we analyzed the S-1 the other day. They told me their long-term compute strategy was to play Google and Amazon off of each other.
(LAUGHTER)
ROOSE: And I was like, that sounds fun. I mean, at the time, you have to understand, like...
NEWTON: Yeah.
ROOSE: ...This was not a real company. They had 150 people. Their office was this cramped little space. There was no security. It was just kind of, like, you take the elevator up, you walk in.
NEWTON: I mean, and in terms of, like, the product - like, Spring 2023, this is, like, sort of very early version of Claude, and usage was probably a rounding error compared to ChatGPT.
ROOSE: Yeah. So the thing I was there to report on was the development and launch of Claude 2, which was their first real public model. So a couple things stuck out to me in going back through my notes.
One was, I was very struck, even at the time, by how sincerely panicked everyone was about AI. I'll just read you some lines from my notes that I took following my first visit there.
(Reading) There is nothing fake or cynical about Anthropic's agita. They are quite simply the most worried company I have covered. I have never encountered a group of people that so badly needed a Klonopin.
(LAUGHTER)
NEWTON: Which is so interesting because, of course, in 2023, that was risible, right? Like, you're afraid of Claude 2? Be serious. And yet, fast forward to 2026, when we're having a fresh outage every day in a new part of the internet for reasons that do seem AI-related, and all of a sudden, the panic doesn't seem so silly.
ROOSE: The other thing that really stuck out to me as I was going back through my notes is I was very skeptical of Anthropic's prospects as a business.
NEWTON: How skeptical were you, Kevin?
ROOSE: So I talked to everyone there, including, at the time, their one business guy. He said that basically the business team at Anthropic - which at the time was, like, him and maybe one other person - essentially serves at the pleasure of the research team.
NEWTON: (Laughter).
ROOSE: He said, quote, "we need to make the money so they can do the work to, like, save the world."
NEWTON: Right.
ROOSE: So essentially the business was very much an afterthought.
NEWTON: The tail was wagging the dog here, a very unusual setup.
ROOSE: Exactly. And I wrote in my notes, they have to make money, but they seem to be doing it begrudgingly. And then this is the line that really makes you believe that I have no future in venture capital. I do not know why you would invest in a company like this.
(LAUGHTER)
ROOSE: They are so worried. They are going to optimize not based on what will make money, but what they can stomach.
NEWTON: You know, it's so funny because saying you have no future in venture capital, it's true and for this reason. Ever since I got to this town, what the VCs always say is that the best investments look like the worst ideas, right? And if you want to make money in this business, you sort of have to bet on the stupid thing. And, man, will Anthropic be an all-time case study for that phenomenon.
ROOSE: Totally. So I was unimpressed by the business, but I was impressed with their seriousness about AI safety and the mission. Everyone I talked to, no matter what division they were in, like, was sort of obsessed with this question of AI safety. I wrote in my notes that they are, quote, "a little weird, but in a "Good Will Hunting" way, not an "American Psycho" way."
(LAUGHTER)
ROOSE: Whatever I meant by that - and it just seemed to me like this collection of incredibly smart, incredibly farsighted people. Then - it's easy to forget - their biggest investor gets hauled off to prison.
NEWTON: Sam Bankman-Fried.
ROOSE: Sam Bankman-Fried. So they are in this period where they are near death, and that's - I didn't know it then because they were projecting sort of confidence and optimism. But this was a company that was sort of, like, on the brink of collapse.
NEWTON: Right. And that was felt within the company.
ROOSE: Yes.
NEWTON: And the company itself was surprised that it did not die on a handful of occasions over the past few years.
(SOUNDBITE OF MUSIC)
DETROW: Casey Newton and Kevin Roose - the new podcast from NPR is called Machine Gods, and we will note that Anthropic is a financial supporter of NPR.
(SOUNDBITE OF MUSIC) Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.