Source: Roger McNamee
I think that Anthropic is a company that shouldn't exist. The issues with the people, the systematic mischaracterization of what they're doing and what's going on in the world. The speculative fiction in their S1.
Roger McNamee was one of Facebook’s earliest investors - and later became one of its most prominent critics.
McNamee began his career running a then-scarce Technology desk at T. Rowe Price in the early 1980s. He cofounded Silverlake Partners in 1999, then left to cofound Elevation Partners, whose positions included Palm, Bioware, and, most of all, Facebook.
His book Zucked: Waking up to the Facebook Catastrophe recounts his path from hope to disillusionment, not just with Mark Zuckerberg and Sheryl Sandberg, but with Silicon Valley’s decaying ethos. The book became the basis of Aaron Sorkin’s new film The Social Reckoning.
A lifelong musician, McNamee still tours, records, and releases music with Moonalice.
You can see the full conversation on the Zero Knowledge YouTube channel. Please like, share, and subscribe!
What convinced you to become one of Facebook’s earliest investors?
Mark Zuckerberg came to my office in May of 2006. He had just had his 22nd birthday. The company had 9 million users. They didn't even have a news feed yet.
I was already convinced that they were going to win in social media because he had two things nobody else had: He gave you control of who could see your data… and because he required you to use your school email address, he had authenticated identity, which meant that the problem that had killed MySpace wasn't going to happen on Facebook.
Why did you walk away from Facebook? What changed?
I worked with Mark for three years, until 2009. One day he walks in to one of our weekly meetings and he goes, 'Uh, we just hit a quarter of a billion users.' And I go, 'Fantastic… Time to monetize.'
He goes, 'No, no, no. I'm going for a billion users… when we get to one, we're going to two. And when we get to two, we're going to three.' And I'm going, whoa, that means you're going to be doing business either under terms you shouldn't want or in places you shouldn't want to be.
I said to him, 'Look, if that's really where you're going, I think my time here is done.’
The Four AI Mistakes
Roger McNamee believes AI firms made four critical errors in their development and rollout strategies.
LLM Technological Focus:
“Large language models, that is not a thing that you can scale. It’s just one [component] that you need to put into a bunch of other things to do anything... It's inherently limited because it's probabilistic, [and] working from human observations of the world rather than the real world stuff."
General-Purpose Architecture:
“When was the last time that a general purpose architecture succeeded in Silicon Valley? It was SAP's R3, the big enterprise resource planning software. That’s 1993, so 33 years ago, and it was a disaster. The average company had to spend at least 10 times as much on the services to make it work inside their systems as they spent on the software itself… Which is why everybody in Silicon Valley has been doing highly focused, rapid payback products ever since."
Spending Big Before Hardware Commoditization:
“These guys have spent $1.5 trillion, and they've made financial commits for additional $3 to 3.5 trillion… They're buying [hardware] at retail and they're saying, oh no it's just like the internet. Dude, no it's not at all. The internet was developed on Silicon Graphics supercomputers, but for commercialization, they waited until they could do it on Sun servers, which cost one-tenth as much. And then when Google came along… they rearchitected everything to work on naked Intel motherboards, which were about 1% the cost… You got roughly three orders of magnitude of cost change in six years and you didn't spend real money until you got to that third level.
“And by the way: investing a trillion and a half before you commoditize this thing means that the trillion and a half dollars is essentially a write-off.”
Letting Ten Competing Players Grow
“There are ten global players who have essentially the same technology, essentially the same training sets, going after the same customers and the same use case. Best case, let's assume the industry really works. How many of those guys are going to be successful? Right. Two. Yep. Maybe a couple more can hang on.
But six guys are gone for sure.”
A lot of the AI industry is controlled by these social groups, like Effective Altruism, that at least profess socially positive intentions.
I view Effective Altruism as a cult. I view the large language model community at large as a group of cults, right? When you talk to people in that world, their perception of the world has been so thoroughly distorted by the experience of being inside that bubble… it's almost impossible to have a rational conversation because they can't understand how you arrive at your conclusions, and it's very hard for us to understand how they arrive at their conclusions.
People don't know the history of AI. Tell us about Eliza.
There was a professor named Joseph Weizenbaum who did an experiment in 1966 with a mainframe computer which he set up the way a therapist works. What he discovered was roughly a quarter of the people developed a really, really unhealthy relationship with ELIZA, which was a text-based bot.
The threshold at which people identify with a machine thinking that it has anthropomorphic tendencies is so low that it was clear in 1966. Weizenbaum spent the rest of his life [saying] whatever you do, do not ever let people make anthropomorphic systems again. And these clowns have gone and done it.
You’ve used the phrase "capital misallocation" to talk about AI. These investments have opportunity costs.
Do you remember the Paris Accords? The Paris Accords on Climate Change said in order to hold climate change at 1.5 degrees centigrade, the world needs to invest $10 trillion over 10 years. The US piece of that was $4 trillion. And we're doing this instead.
How about healthcare? How about housing? How about education? How about dealing with the psychological problems of young people? I mean there are so many things you could literally change from being problems to being benefits with small fractions the amount of money we're talking about.
OpenAI coming up short on fundraising, Nvidia's stock crashing 20%; does this crush the Anthropic IPO?
We can only hope that it does. I mean, I think that Anthropic is a company that shouldn't exist. You know, that the the issues with the people, the systematic mischaracterization of what they're doing and what's going on in the world, the speculative fiction in their S1.
This company is deeply unprofitable. Here's the simple question: Can anyone identify an enterprise customer who pays more for AI than it costs the vendor to provide that service?
If such a person existed, there would be white papers about them everywhere.
It would be in the interest of everybody to promote that.
- Roger McNamee
Thanks for reading. We welcome suggestions for future interesting people to talk to.
Managing Editor David Z. Morris
