I can’t believe this is happening transcript

I resigned from Anthropic today. I spent the last 3 years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below. Now, this tweet's got 172 million views. That is the biggest tweet I have ever seen, not just in AI safety, but in all of Tech Twitter. This is the most insane thing I've ever seen. 800,000 likes, 200,000 re posts.

This is by far the largest Tech Twitter experience of all time. In fact, Bernie Sanders tweeted about it. Dario did. Sam did. Elon did. A kernel engineer from DeepMind called Anthropic Hitler. Trump ended up calling Jensen on stage to talk about how we need to accelerate and not listen to Dario. And the psyops came out. Is this really Dario just trying to take control with his cabal of friends to be able to control the internet?

Did I create this graph so that I can give my best Charlie Day impersonation? Yes, I did. Is there a movement called pause AI where everybody looks sad? Yes, except for this guy. Why are you smiling? Now, I know a lot of you aren't on the internet, and so I'm going to walk you through what happened, and then we're going to jump into the conspiracy theory. Is Meter controlled by Dario? Is it just an Anthropic front, and is this a play for regulation?

Because you know me, I like I I like myself a good conspiracy theory. But of course, before we get started, a quick thank you from the sponsor. The next 100 billion commits are coming faster than anyone has expected, and that means that your CI it needs to be blazing fast. Namespace optimizes for fast coding, testing, and build environments across Windows, Linux, and Mac. Mac. They even have Mac? That is their singular focus.

And that focus runs down to the metal because they design and deploy their own server racks around the world which runs their GitHub action runners. You can remote desktop or even SSH in, and you know I love SSH and debug why a test is failing. You or your agent can optimize performance issues. Namespace is trusted by 11 Labs, Ramp, Warp, Vanta, Framer, Zed, Fowl, Sierra, Whisperflow, DuckDB, even Ghosty uses them. You should try it out today at namespace.so.

All right, so this story centers around a AI researcher named Jason Coxen. Coxen, Coxen, Coxon. I'm not going to make that joke. AKA mini Dario, Dario 2.0, or Harry Coxon. Again, not going to make the joke. On September 8th at approximately 6:04 p.m. the Lord's time, Montana time, he made this tweet that went omega viral. 18 minutes before that tweet, Wall Street Journal publishes this article referencing Jacob. Shortly after Evan, a alignment science lead at Anthropic, retweeted his tweet and said Jacob here is correct AI giveaway.

We really do earnestly believe AI could kill all humans. I personally think it's greater than 10% chance within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track, too. Now, I know some of you when you hear that you're like, "Oh my gosh." Well, guess what, buddy? Just 1 year ago today, Dario actually said we had a 25% chance for things to go very, very badly.

So, you know what that means? Things are looking up, boys, okay? My life is like stonks right now. Humanity, stonks. Another alignment team lead at Anthropic said Jacob was a senior researcher who joined Anthropic in May. My Anthropic colleagues and I have been trying to recruit him for over 2 years because we knew he was a strong researcher at OpenAI. Before he left, I pitched him to stay and join my team. I was sad when he decided to leave as are many of my colleagues. 100% agree with him that AI poses serious risk to society and I'm glad he's speaking out.

Well, this might prove to be the most important whistleblower ever. Anthropic's CEO just said, "It's my worry that in 6 to 12 months such a swarm could be capable of taking over the entire internet." Only 4 days after the tweet, Jacob might literally save the internet. Now, of course, when Dario saw the omega success of the tweets happening in the support for Jacob, he released the article "We Must Pace the Frontier" in which he proposed a three-step plan to help pacing the frontier.

And the most important one for our discussion today is number one because this is really where all the conspiracy theories just just explode at, okay? So, embedded evaluators. Each frontier AI company commits to giving ongoing employee-like access to a team of embedded third-party evaluators such as Meter. Keep Meter put in the back of your brain right now, okay? Whose role is to verify adherence to safety practices and commitments, report incidences, and help assess the alignment of not just completed AI models, but training pipelines and processes.

In other words, along with regulation by the government, Dario hopes to be able to create these embedded evaluators such that they could determine if an AI lab is really worthy to pursue superintelligence. And so, one could imagine that these third-party evaluators are going to hold a lot of power in Dario's future. Now, Sam, I don't really like to hold hands with Dario. Altman, he actually tweeted he agrees with Dario.

Now, I don't know if this is the first time, but this feels a little unusual. Sam agreeing with Dario? Also, Elon Musk agreed with Dario. So, naturally, when you see all these political figures come together along with Sam and Elon and Dario all agreeing that we must pace the frontier, this is naturally going to raise a little bit of suspicion that this all just came together in a moment. And it first started with, "Man, that tweet it really just exploded. 24-hour check-in, 123 million views, 656,000 likes, 190,000 followers.

I've never seen any AI safety tweet go this viral before ever." Even Elon Musk goes, "Yeah, I've never even seen an account go that viral before with no interactions or no followers. And what he means by that is this account, Jacob Coxen, when you go to the about for this account, it was created this year, it's changed its name twice, and there's only two tweets. The tweet he originally sent and an ask me anything which happened approximately 7 days later.

That's it. That's all this account has done. In fact, his oldest reply is to his first tweet. Some would call that unusual. Second, the whole Wall Street Journal thing, how did the Wall Street Journal know about your tweets before they came out? exclusive access to journalists That just That just is Okay, hey, you know what? I don't I don't want to be a conspiracy boy, but I'm just saying that is so strange. I have never went Okay, I'm going to quit a job, but hey, before I quit this job, I got to reach I got to reach out to the news.

I'm about to talk some about my employer. Like that is That's a That's a new one. Now, I get it. His goal was to go out there and talk about AI safety, so he wanted his message to go as far as possible. At least that would be the charitable read, but it is just a little strange, right? It also actually gets stranger because Netflix just happens to have a documentary coming out about the dangers of AI, and it took 3 years to make and just happened to come out on the same week.

Also, Time magazine just happened to have their cover planned this week for you guessed it. Pausing AI. How dangerous is AI? Now again, I'm not a true believer. I'm not actually a conspiracy guy. I just love going through all the motions, but I will say that man, if you wanted to make somebody a true believer, this is really making me feel a certain way. But here's the thing. It actually gets a lot worse. In fact, it gets it gets a little ridiculous.

So, I want you to hold on to your butts because we are about to go all in on some conspiracy theories. All right, so to get started on this whole theory that METER is this organization in which is trying to create a cabal to be able to rule over superintelligence and it's being backed by effective altruists, EAs. Uh you have to start here first. Remember Dario's article when I told you, "Hey, put Meter in the back of your head." He names a singular embedded evaluator, and it was Meter.

So, I got to show you this. But, before we do, there's only one thing I can do at this point. I need a second tin foil hat. Like, if we're going to go here, honestly, I feel like I just don't have enough tin foil hatting. So, this is Jacob Coxen. He's the one who kind of launched off this entire thing with effective altruism, Meter, pause the AI, everything. Everybody started looking at this cuz of this guy. Now, a very important character in the effective altruism kind of universe is Dustin Moskovitz.

He has something called Good Ventures, and he also has something called Open Philanthropy. Good Ventures sometimes funds Open Philanthropy. There's another character named Holden. He also helped co-found Open Philanthropy. All right, let's introduce Jan, John, Yan. He created something called Survival Flourishing Fund. And let's introduce one more major character named Paul Christiano. He created something called Arc.

You're probably familiar with the Evals. They're like, you know, can't is it really AGI Evals Arc. Now, what connects Coxen to all of this is that he actually took about $20,000 from Good Ventures when he was younger. Now, this hasn't been explicitly confirmed by him. It's just all over the internet. He also was a fellow at New Speak House in London, which is an effective altruism guild, like a group. All right, so let's bring in the big man, Dario.

Okay, now we're going to put this all together, people. We put Anthropic in the middle. Now, he founded Anthropic because he believed OpenAI wasn't the right place to shepherd in super intelligence. Daniela, his sister, came to work with him at Anthropic, and yes, is his sister. Now, Dustin Moskovitz was an early investor in Anthropic. Holden and Daniela, they got married. Fun fact, by the way, Holden, Daniela, and Dario, while they're starting Open Philanthropy, all lived together.

Just fun fact, just in case you're wondering. Daniela and Holden were in fact engaged at that point. Holden is an early investor in Anthropic. Meter was spun out of Arc. Now, this is very important, okay? Meter was spun out of Arc, and Meter has taken about 200 and some thousand dollars from Open Philanthropy. Now, there is a person named Ajeya, and she works at Meter. Keep her in your mind for a second, okay? Paul Christiano and Dario, they also used to live together.

Just fun fact. Obviously, OpenAI, Dario used to work at OpenAI, left OpenAI. Holden also worked at OpenAI. Anthropic spun out something called LTBT, known as the Long-Term Benefits Trust, in which Paul is the trustee. Paul and Ajeya are also married. And Ajeya used to work as AI and safety at Open Philanthropy. Meter has received over five million dollars from Survival and Flourishing Fund. At least those are the numbers I found all over the internet.

Jan, who started Survival and Flourishing Fund, also early investor in Anthropic. Now, all of these guys are EA guys, but you can't really be an EA group without the big man, Sam Bankman-Fried, okay? EA is effectively Sam's morality. And Sam, he founded FTX, and FTX also gave over a million dollars to Meter. And Sam also is an early investor in Anthropic. I know, can you believe that, man? Sam Bankman-Fried, early investor in Anthropic, early investor in Cursor.

I mean, this guy should have been a billionaire 10 times over. It's crazy that he's in prison. Like, what a crazy world we live in. And then, of course, the other connection for Jacob is that he worked at Anthropic, and he worked at OpenAI. And so, that's kind of what ties him into all of this. Now, if you really just break it down into Meter, Meter takes money from Arc, SFF, Open Philanthropy, FTX, and has Ajeya working there.

When you expand that out, that's Dustin Moskovitz, Sam Bankman-Fried, John, Holden, Paul Christiano, all together. They all have strong connections to Dario. So this is the entire argument. And this is exactly what David Sachs was arguing about. Stop pretending you need anyone else's permission. Anyone else's permission being like you want to slow down or pace the frontier. Stop pretending antitrust law has to be suspended so you can form a cartel.

Dario does suggest that all the AI companies should be able to work together to determine how AI should roll out. Stop pretending you need regulatory approval process that supersedes product liability. Stop pretending Meter isn't independent when it is intertwined with Anthropic's investors and staff. Stop pretending you need those same evaluators to police competitors who aren't even at the frontier. And so this is the whole thing.

This is the the big conspiracy theory that I wanted to talk about, which is that Dario is making a play to control everybody. He's ultimately felt like he is the one that should shepherd in superintelligence. He wants to be everybody's dad. We have said that. And this is the play. This is how he's going to be able to do it by having third-party investors that are actually strongly tied to him and his thought process.

So he will be able to say no, you can't and yes, you can based on these third-party evaluators. So what are my thoughts on this? Do I think that Dario wants to control AI? Yeah, absolutely. I mean, he's has so much said that he believes that he is the person that should be the one shepherd again. He thinks Anthropic is doing everything right. Well, he thinks every other competitor is being dangerous and acting irrational.

Do I think that he wants to be able to have regulatory stuff on AI? Yeah, he said all these things. Does he have a lot of connections with Meter? Yeah, he probably has a lot of connections with Meter. If he were to talk to people at Meter, would they actually listen to him? Yeah, probably, right? Like this all of this just makes sense. Is it some grand cabal? I don't know. That's that's impossible to say. But I wanted to tell you my real thought process.

You know what my conspiracy theory is? Remember when Sam agreed with Dario that we should pace the frontier? And remember when Elon agreed with Dario that we should pace the frontier? My firm belief is that Elon and Sam, they're gassing up Dario. They're like, "Yo, dog. yeah, absolutely, absolutely, we need to kind of, you know, slow things down, okay? We need to pace the frontier." They're not going to pace the frontier.

Do you know for a fact right now they're like, "Oh my gosh, I think he's buying it. I think he's buying it. I think he's going to police himself." And they're going straight to the moon. All right, well, that's the internet. Hey, hopefully 90% chance, huh? Like, I mean, that's pretty good odds. Like, that's higher than a lot of my grades in high school and uh higher than most my grades in high school, in fact. Okay, bye-bye now. >> I'm on my knees, bloodied by an anthropic killbot, pushes a gun into my temple. >> Any last words? >> Coxen was right. >> Coxen? >> I grin through my broken teeth. >> Cox on your face. >> Arjen. >> All right, I actually have to do a bonus ending right here.

It turns out that Jacob said he worked with no third parties to set up his resignation live on TV. Well, it turns out he did. Not only did he say it himself, but in the Wall Street Journal, it also shows that he worked with people that are heads of policy at anti-AI policy places before his resignation. Turns out that Jacob just lied a whole bunch.