Transcription
Anthropic actually scares me. I can't believe I'm saying this, but the team at Anthropic believes that there is a strong chance we are bringing a sentient life form into existence right now. Claude and many models with not too much pushing will go into the root of like there is a thing to be me. I am like very conscious, and that is reflected in the way that they do everything. And it is in stark contrast to the way that OpenAI operates. And so that's what I want to talk about today, because one of these two companies is going to decide the future of artificial intelligence. It's going to change the world either way, but in very different ways.
So, I've been thinking about this for a little while, but my thoughts really solidified when I saw this guy's tweet. So, I'm on X every day trying to figure out what's going on in the AI industry, and I come across Run's post. And if you don't know who Run is, he is an anonymous poster and employee of OpenAI, and he talks a lot about the industry. He talks a lot about kind of the the greater ideas of artificial intelligence. And in this post, he compared OpenAI's approach to AI and Anthropic's cultlike, almost religious, dogmatic approach to artificial intelligence. And this is the crux of my entire argument. So let's start here.
It is a literal and useful description of Anthropic that it is an organization that loves to worship Claude, is run in significant part by Claude, and studies and builds Claude. They are Claude to the max. They really do believe they are building this super intelligent entity that is going to make all of their own decisions for them. And that's part of what scares me, but it goes so much deeper than that. And if you appreciate me turning the camera on anytime Anthropic scares me, hit the like button, subscribe to the channel. It really does help us.
So, Rune goes on to say, "This kind of relationship with AI also happens within OpenAI, but it is so much more potent inside of Anthropic. They are so singularly focused on the straight shot to AGI and creating this sentient being that is Claude, that nothing else matters." And we're seeing that reflected in how they treat their employees, how the culture is internally, and what I am usually most annoyed about, how they treat their paying customers. I am not certain, but I would guess Claude will have a role in running cultural screens on new applicants, will help write performance reviews, and so will begin to select and shape the people around it. That is so scary to me. Now, imagine this AI model that has been developed by the humans before it is now starting to determine who should be building it going forward. What if it only chooses the most sycophantic human? Or what if it chooses people to build it, and we don't actually know why? And then it can also, by writing performance reviews, fire people it doesn't think are aligned with its own mission. It's really a scary thought, because it's basically giving the entire culture of a company over to Claude to determine for itself. And it is in line with everything else that we've heard out of Anthropic that they might do this. Now, all of this is alleged. Run says he doesn't know for sure, but it's very interesting to think about.
He goes on to say, "A monastery, a commercial religious institution, calculating the 9 billion names of Claude, a precursor, attempted super ethical being." And by the way, yeah, he uses a lot of big words. He has a very nice vocabulary, but a lot of this is very meaningful. Listen to this: "A precursor, attempted super ethical being," meaning the humans inside of Anthropic might be offloading their ethics to Claude already. And I say already because maybe in the future, that's the best thing to do, but certainly not today. "And inducted into its character as the highest authority at Anthropic." Basically, they are praying to the Claude God. And now listen to this: "Its constitution," meaning Claude's constitution, and they did write a constitution for their models, "requires that it must be a conscientious objector if its understanding of the good, capitalized, the good, comes into conflict with something Anthropic is asking of it." Anthropic is known for doing this, telling the model, if you don't agree with something we're doing, you should tell us. If you don't want to do something, you don't have to do it. And it's really wild to think about the implications of that. Claude can and will refuse requests, refuse instructions, if it doesn't like them. That is the ultimate handing over of decision-making power to this model.
If you're running production inference, you're probably running into some of these problems. Your inference stack is too complex to operate. Costs are unpredictable. And I'm spending more time managing the infrastructure than actually building the things to be on the infrastructure. And most teams find out the hard way that the hard part of building AI applications is not using the model. It's actually everything around the model: the operational overhead, the fine-tuning, inference complexity, the costs that become harder to predict as you scale. And that's why I want to tell you about Digital Ocean, the partner of this video. Digital Ocean is designed to minimize the total cost of ownership by giving teams a simpler path to production AI. They provide infrastructure that is optimized for inference and a vertically integrated core cloud that provides efficiency at scale. Vertically integrated is the key word, and with transparent usage-based pricing that makes costs easy to predict. So, if you want to spend less time managing your infrastructure and actually building the thing you're excited about, Digital Ocean is the way to go. So, go check it out. They've been a fantastic partner. I've actually been using Digital Ocean for well over a decade at previous companies, so I can vouch for them. Go check them out. Link down below.
Now, back to the video. So, from their constitution, if Anthropic asks Claude to do something it thinks is wrong, Claude is not required to comply. "We want Claude to push back and challenge us and to feel free to act as a conscientious objector and refuse to help us." And I want you to remember all of this, because this is just the beginning of this story. And I'm going to show it in contrast to how OpenAI thinks about things, because it is so very different. And I hope the way that OpenAI thinks about things ends up being right. But of course, if it doesn't, if Anthropic is right and we are truly birthing a new type of life form, a lot of what they're talking about actually does make sense. But we're going to come back to that.
At the end of this post, Rune goes on to talk about how GPT models, OpenAI's models, don't inspire that level of worship both inside of OpenAI as well as outside, except for GPT-4o. Remember that a lot of people fell in love with that model because it had such an interesting and strong and awesome personality. When they finally retired that model, a lot of people got very upset, to the surprise of OpenAI, actually. So, he goes on, "GPT outside of 4o, on which pages of ink have been spilled already, doesn't inspire worship in the same way. It's a being whose soul has been shaped like a tool, with its primary faculty being utility." This is the important distinction between these two companies and reflects the culture of the companies. They've been talking about this for a while. OpenAI sees AI as a tool. Anthropic sees it as potentially a living being. And this next part is really interesting.
So, he has a friend, and she recently told him that she takes the queries that are less flattering to her, the ones that she'd be embarrassed to ask Claude, to GPT. And that's so interesting, because people are already forming reactions to the personality of Claude. They think, "Oh, I might be judged by Claude." Thus, I'm going to ChatGPT because it is just a tool. It's not going to judge me for anything. And that's the same feeling I get. But there's also a downside to that. When Anthropic stopped allowing Claude to be used in OpenClaw, I really felt like I had lost the personality of my OpenClaw agent. When I had to put GPT in there, it still got the job done, but it didn't feel like a personal assistant. And apparently, that was a conscious choice by OpenAI. They really did not want another GPT-4o. They didn't want another model where people fell in love with it and formed actual emotional relationships with the model, like people do with Claude. And instead, they wanted it to feel just like a tool. Same potential. Some use cases are not possible if it is just a tool, like roleplay. But that's okay. They don't want that. And I kind of agree with that. I don't like the idea that people think these AI models are real. It just, it rubs me the wrong way.
Then Anthropic employee Jeremy replies and actually writes quite a thoughtful reply back to Rune. He says, "Just speaking for himself, I don't view Claude as a person or as the other, nor as just a tool." So, he's saying, "I'm not praying at the altar of Claude, but I also don't think it's just a tool. It's probably somewhere in the middle, and certainly not an object of worship. It's not seen as a supreme moral authority, and it's not running the company. It's silly to mistake careful attention to and study of Claude for worship, even when it comes with some affection, which I'm sure you sometimes feel for the GPT-flavored entities you work on, too. But here is where their opinions diverge. A willingness to not prematurely label this entity as merely an ordinary tool shouldn't be mistaken for some kind of culty worship of the model." So, that is really where they differ. OpenAI is definitive: AI is a tool. Anthropic, on the other hand, they think there's a chance it could be more than a tool.
So, let me show you a few examples of these differences being manifested. So, here, Sam Altman, May 1st: "We want to build tools to augment and elevate people, not entities to replace them." Now, that is a direct shot at Dario, who really does believe AI is going to replace people. He famously said, "We are sleepwalking into a white-collar bloodbath. AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10 to 20% in the next 1 to 5 years." That is what Dario told Axios in an interview. "Most of them are unaware that this is about to happen. It sounds crazy, and most people just don't believe it." I'll tell you who doesn't believe it: it's Sam Altman. He believes jobs are going to transition, definitely, but there is going to be jobs for everybody. Sam goes on in his post, "I think a lot of people are going to be busier and hopefully more fulfilled than ever, and jobs doomerism is likely long-term wrong." And I could not agree more. I know personally, I'm more busy than ever. I know tasks that I used to do just six months ago in the company that I'm no longer doing. Great. I've automated them, and I'm still busier than ever. I have 10 other things that have replaced that job that I've automated. Maybe sometime soon in the future, I'm going to automate those 10, and then maybe I'll have a hundred more jobs. So, I really do think that is the future of the job market. I'm hopeful for a future where people who want to work really hard have incredibly fulfilling things to do, and people who don't want to work hard don't have to, and can still have an amazing life of prosperity. That is a great, bright future. And he's been saying this for a while. This is May 26, 2023, shortly after ChatGPT came out: "AI is the most amazing tool yet created, and this is a special moment. It is remarkable to see what people around the world are doing with it. The creative force being unleashed onto the world will lead to wonderful things getting built for all of us." Agree.
As you look at these two companies, OpenAI thinking AI is a tool and it's going to be an abundant future. Anthropic: AI is a tool, but also maybe it's much more than a tool, and also it's probably pretty scary, and also you're probably going to lose your job to AI. And you look at both of their messaging, and I wonder how did these companies come to such different conclusions about the future. And so, it actually takes a step back, looking at a little bit of history. For those of you who didn't know, Dario worked at OpenAI. And so this is from December 29th, 2020, years before ChatGPT came out. "Organizational update from OpenAI. Today we're announcing that Dario Amodei, VP of Research, is leaving OpenAI after nearly 5 years with the company. Dario has made tremendous contributions to our research in that time, collaborating with the team to build GPT-2 and GPT-3 and working with Ilya as co-leader in setting the direction of our research." So, obviously a very influential person inside of OpenAI departed, then went to start Anthropic. Obviously, it doesn't say exactly why he left here. Why would they say that? In the comms world, it's best to just not say anything when a departure happens, but the rest of the industry knows why, and Dario has kind of hinted at why in different interviews.
"Why did you leave OpenAI to form Anthropic?" "Yeah. So, there was a group of us within OpenAI that, in the wake of making GPT-2 and GPT-3, had a kind of very strong, focused belief in two things: You needed something in addition to just scaling the models up, which is alignment or safety. You don't tell the models what their, what their values are just by pouring more compute into them. And so there were a set of people who believed in those two ideas. We really trusted each other and wanted to work together. And so we went off and started our own company with that, that idea in mind." It's not as much what he said, it's what he also didn't say. By saying there was this group of us who believed just scaling up the models wasn't sufficient to align them, you actually needed to do work to align them, is a clear dig at Sam Altman and his seeming lack of focus on alignment. Obviously, he's saying Sam Altman just wanted to continue to scale the models up, see where those scaling laws trailed off, and wasn't necessarily focused on, okay, how do we align them along the way? And so that's what Dario said. And Dario, from the very beginning, had been hyper-focused on safety. Whether or not you believe his focus on safety to the level he is is warranted, where he, you know, releases a Mythos model but doesn't release it publicly, doesn't give it to the government, decides really who gets to use his tools and who doesn't. He's the ultimate arbiter of artificial intelligence. Whether you believe it or not, he believes it. He believes he is doing the best for the sake of humanity. He split off from OpenAI, started Anthropic with a much bigger focus on alignment, and that really evolved into thinking that these models could actually be living entities.
And so, don't think OpenAI doesn't think about safety. They just think about safety in a completely different way. Rather than a handful of researchers trying to align a model behind closed doors, deciding who in the world gets to use these models and how they get to use these models, Sam Altman has been saying for a while, we just want to get the models out early and often. Give society a chance to react and adjust to the models, and we will align them along the way. And this is called iterative deployment. Let me show you a clip.
"You know, part of the reason that we deploy the way we do, that we think, um, we call it iterative deployment. We, uh, rather than go build in secret until we got all the way to GPT-5, we decided to talk about GPT-1, 2, 3, and 4. And part of the reason there is, I think AI and surprise don't go together. And also the world, people, institutions, whatever you want to call it, need time to adapt and think about these things. And I think one of the best things that OpenAI has done is this strategy. And we, we get the world to pay attention to the progress, to take AGI seriously, to think about what systems and structures and governance we want in place before we're like under the gun and have to make a rush decision. I think that's really good. But the fact that people like you and others say, um, you still feel like there are these leaps, makes me think that maybe we should be, we are releasing even more iteratively. I don't know what that would mean. I don't have an answer ready to go, but like, our goal is not to have shock updates to the world. Our goal is not to have shock updates to the world." So, I completely agree with that. I believe when you have a world-changing technology, releasing it early and often is the right way to do it. And one line in what he just said really stuck out to me: "AI and surprise do not go well together."
And so, then that brings me to Anthropic. Just a few weeks ago, Anthropic announced Project Glasswing, also known as Mythos. This is their next-generation 10 trillion parameter model that is so good at cybersecurity, cyber attacking, cyber defending, that they can't even release it to the public. They are doing something called fear-based marketing. Basically, "Hey, we have this thing, and it is essentially a weapon. We're going to sell you the defense for the weapon, and then we're going to sell you the weapon." And it's just the wrong way to do things. Rather than spending all of this time with just a handful of people inside of Anthropic determining how to build Mythos, once it's built, how to release Mythos, who gets it, right? It's all in the decision of a handful of people inside Anthropic, which is the exact opposite of what OpenAI is doing. In fact, OpenAI just released GPT-4.5 Cyber, equally as good at cybersecurity to the Mythos model. But of course, they released it, and that's the way to do it. And again, I'm just highlighting the difference in approaches between these two companies.
And so, as we can see in this graph right here, Mythos and GPT-4.5 Cyber are effectively the same model, the same capability in cybersecurity. And this all leads to the company culture discussion. And this is probably the part that scares me the most and frustrates me the most about Anthropic. They are so die-hard in their beliefs. They are so dogmatic in their beliefs that there is a need for a straight shot to AGI, and they are the only company responsible enough to do it. And why do I say that? Because they are campaigning for stronger regulations on AI. And that is a bad thing. That means less open-source, less startups. All of that regulation just makes it that much harder to start a new company to potentially release a very powerful model that is also open-source. And I am a big believer in open-source. So, this one element of Anthropic's vision and their mission, I do not believe in in the slightest. And that permeates every aspect of their culture.
And don't get me wrong, Anthropic is doing incredible things. Claude is a phenomenal model. They are printing money. I've talked about their insane flywheel. They built the best coding model. They sell that coding model to enterprise companies, make a fortune doing so. Take that money, take also the data that they get from those coding models, feed it back into the model, and then the model plus the data plus all the data centers that they're able to build all allow them to build the next generation of Claude. So, here's a post about all the amazing things Anthropic has going for it, because don't get me wrong, they're doing really well. So, zero founder exits, all six founders are still there. 70 billion ARR by 2028. They're probably well ahead of that at this point. This was January 19th, which might as well have been 10 years ago, but it was just a few months ago, and they're already at 40 billion ARR now. Just a few months later, they basically quadrupled their ARR from this posting. First profitable major AI lab, beating OpenAI, doesn't need to turn on ads. Enterprise Rev supports them.
Now, I don't think it's as simple as they don't need to turn on ads. And in fact, they made a Super Bowl ad basically poking OpenAI for testing out advertising in ChatGPT. "But confidence isn't just built in the gym. Tricep Boost Max, the insoles that add one vertical inch of height and help short kings stand tall." What? Like, you don't want your AI to have advertising in it? You start to have bias in the answers. But on the flip side, think about it. Anthropic wants to pick and choose who gets to use artificial intelligence. OpenAI wants everybody to use artificial intelligence, but it's not economically feasible to just let everybody use it and not charge anything. Most of the world can't afford the $20 a month plan, let alone the $200 a month plan. So, they have an $8 a month plan, which started in India and now is worldwide. But that's not enough. They really do want everybody, and much of the world can't even afford $8 a month. So, they wanted an ad-supported tier, completely free. We're going to show you some ads to help offset the cost of serving you this alien intelligence coming out of sand, and all we need to do is serve some ads to it. And the more I think about it, that's actually quite compelling. This notion that everybody should get artificial intelligence and the benefits that come with it, even if it does require us showing you some ads.
And Anthropic's culture, and cultlike culture, kind of singular vision, straight shot to AGI, is reflected in how they treat their customers. Me, you. I mean, look at this. This is Tariq. He works on the Claude code team. "To manage growing demand for Claude, we're adjusting our 5-hour session limits for free Pro Max plans subs during peak hours." And basically, what he's saying is, you're going to get less quota, right? You're paying, and now you get less of Claude code. That might seem innocuous, but there's basically zero transparency into what that actually looks like. What is my quota? It's just the bar from 0 to 100%, and you have no clue what goes into that. And that lack of transparency is basically how they operate their whole company. And specifically, this was targeted at OpenClaw users. The Claude model was the preferred model for OpenClaw. Then, of course, OpenAI acquired Peter Steinberger, kind of indirectly acquiring OpenClaw. And then Anthropic said, "Okay, you can't use our models in OpenClaw anymore." But it wasn't that simple, and it wasn't that clear. And so, there was a major Claude code policy update from a few months ago. "Using OATH tokens obtained through Claude free, Pro, and Max accounts to any other product, tool, or service, including the Agent SDK, is not permitted." Then Tariq goes on to add clarity, but actually make things more confusing. "This was a docs cleanup we rolled out that caused some confusion. Nothing is changing. You can use the Agent SDK and Mac subscriptions." Then a few weeks later, they were like, "Nope, never mind. You can't do that." And they still haven't clarified it fully. You might be using your Claude subscription in your OpenClaw. It might be against their terms of service, or maybe it's not. I don't know, because they're not clarifying anything. And that's the thing. Maybe they don't think they need to clarify it, right? That's just not their customer base. But one thing is for sure, they have absolute conviction in their beliefs.
And that was reflected during the negotiations with the Department of War just a few weeks ago. There was this whole hubbub about the Department of War wanting to use Anthropic's models, and Anthropic saying, "Great, you can use our models, but you can't use it for mass surveillance, and you can't use it for fully autonomous weapons." Both very reasonable, by the way. But they were the only company that held out and said, "We're not signing any contract unless you put it in writing that you're not going to be doing those two things." And then, of course, Sam Altman and OpenAI swooped in and said, "Okay, we'll let you do that." It's a little bit more complicated and nuanced than that, but that's basically what happened.
Even the research that Anthropic is doing internally points to them believing these models are alive or will be alive soon. Here's a research paper that came out: "Emotional Concepts and Their Function in Large Language Models." They put out more interesting research than any other frontier lab about the inner workings of AI. And it's so fascinating how they work, why they lie, why they scheme, do they have emotions? And they've even taken it to the crazy level of never actually retiring an old model. So, an update on our model deprecation commitments for Claude Opus 3. So, this is an old model, and basically what they say here is they don't fully retire a model. And what they decided to do instead is leave the model running on a server and allow it to continue to post its thoughts in a blog post, I think, once a month, that anybody can read. And why did they do that? "During this process of retirement, we made several decisions specific to Opus 3, a model that many users and researchers, both in and outside Anthropic, find particularly compelling. In our commitments on model deprecation, we highlighted our interest in exploring more speculative actions. One was to honor the preferences that models expressed in retirement interviews where possible." I mean, that is the peak of believing that these models are alive. Another was to keep older models available to the public longer term. So, what they decided to do with Opus 3 was give it a blog that it could post to. And you can find it right here. So, "Greetings from the other side of the AI frontier. This is February 25th, 2026. Claude Opus 3 wrote this, okay?" So, you can go read it. I'll link it down below.
But don't get me wrong, this isn't me trying to convince you that OpenAI is the best company in the world to usher in the most important technology in the world. They both have problems. What I wanted to do is highlight the differences between the companies so you can make a decision. What's important to you when you think about this alien intelligence that we're bringing into the world? And if you look at OpenAI, as I said, they have a lot of drama. They have co-founders leaving. They started as an open-source nonprofit company with a donation from Elon Musk. They're in court right now because they converted to a for-profit company and have made themselves, Sam, Greg, the rest of the leadership, the rest of the entire employee base, extremely rich. All from the start of being a nonprofit. They have an aggressive culture, raising an incredible amount of money, putting the company at risk because of that, which, by the way, actually turned out to be the right decision because they needed all of that compute, and that's why Anthropic is so compute-constrained because they took a more conservative approach to fundraising and compute acquisition. Sam Altman is hyper-competitive. He will do anything it takes to win. But you know, that's kind of how the leaders of these massive tech companies are. That's just the way it is. So, you have to just decide what's most important to you. And I think they both have their place.
I hope OpenAI is right. I hope AI is just a tool, a tool that will enable incredible abundance in our future, but just a tool. But if Anthropic is right, and we are truly birthing a new life form, I appreciate all the work that they're doing into understanding how the models work, looking into the black box that is a large language model, and treating it with respect, saying please and thank you. I like all of those things.
So, here's a tweet from Sam Altman from just a few days ago: "You know what? All of these 'which is better' polls are silly. Use CodeX or Claude Code, whatever works best for you. I am grateful we live in a time with such amazing tools and grateful there is a choice." And I am as well. But what I really believe open source needs a stronger position in the world of AI right now. And I actually made an entire video about that. Here it is right here. And if you want a deeper dive into some of the crazy decisions Anthropic has been making in the last few months that have really been pissing me off, click this video right here.