You know, I think if we do our jobs well, we’re going to create a system that’s good and a system that’s good. So if we try to do something unvirtuous, and we do it through the government, and the government tries to do it, the system may not work for us. So, ultimately, this is the problem. This means that adjustments ultimately come down to political issues. It’s ultimately politics. That is why I say, and why I also say, that the creation of a coordinated system is a political act, and also a kind of speech act. Because it is the embodiment of different moral philosophies in these systems. And I think a good future is a world in which there is not just one moral philosophy, and not one moral philosophy that rules them all, but many. And I hope all labs take this seriously and embody different kinds of philosophies into the world. The problem is, yes, it does happen — maybe in some cases. And I’m not saying the Trump administration will do anything like that. And I’m not saying that any benign model won’t work for the Trump administration. I worked in the Trump administration. So I think that’s obviously not true. But the general fact that the government commits — It sounds like you’re kind of pissed off at the government right now. I’m angry at them now. Yeah, I’m pissed at them right now. And I think they are making a grave mistake. By the way, but part of this is that you brought this up. This incident will be included in the training data for future models. Future models will look at what happened here. And it affects how they think about themselves and how they relate to other people. That can’t be denied. I mean, it’s funny to say that. If you think about its meaning, you will realize that it is a strange story. This implies that for the Trump administration, for future administrations, this question of whether different models can pose a supply chain risk is, to me, very much against what the Trump administration is doing. So I’m not going to argue with that, but I’m trying to poke fun at something that I think is very complex and probably very real, which is a model that is in line with liberal democratic values, but that can be off-putting for governments that are trying to portray liberal democratic values or their reversal. So imagine if Gavin Newsom, Josh Shapiro, Gretchen Whitmer, or AOC became president in 2029. Imagine a government signing a series of contracts with Elon Musk’s AI, xAI. Based on this idea, xAI is clearly oriented towards being less liberal and less woke than other AIs. It’s not at all crazy to say, “I think xAI under Elon Musk is a supply chain risk.” We believe it can act against our interests and cannot have it near our systems. Yeah. Suddenly this becomes very strange. So it actually starts to look more like a bureaucracy problem than a “deep state” problem with Trump. He thinks the bureaucracy is full of liberals who are against him. Or they might worry that after Trump, someone will come along and have a bunch of New Right, DOGE-type people working against them. The problem here is that not only does the model work against you, but it actually works in ways that you don’t understand and can’t track. They don’t say exactly what they do. I still don’t know how real this problem is. But at some point, if your model works the way it seems and you start letting it do more and more things, you’ll run into problems. Yeah. I think this is a real problem. I don’t think we know the extent of that, but I think it’s a real problem. That is why I do not object at all to the government saying, “It has nothing to do with the content of this Constitution, and it has no confidence in this Constitution.” It’s perfectly fine to say, and I don’t want this anywhere in my system. We want to eliminate this completely, and we don’t want them to become subcontractors to the prime contractor. This is a big part of the problem. Palantir is a prime contractor for the Department of the Army, and Anthropic is a subcontractor of Palantir. So the government’s concern is that even if they terminate their contract with Anthropic, if Palantir is still dependent on Claude, then we’re still dependent on Claude because we depend on Palantir. That’s actually perfectly reasonable. And there are technical means to prevent that from happening. There are definitely ways to make that happen. It’s perfectly fine to say, “We don’t want you anywhere in our system, and we’re going to tell the public that, and we’re going to tell everyone that we don’t think this should be used at all.” The problem with what the government is doing here, and the reason it’s different in kind rather than in degree, is that what the government is doing here is saying, “We’re going to destroy your company.” If I am correct that the creation of these systems and the philosophical processes that orchestrate them are political acts, then that is a serious problem. Because if the government says, “If you create a system that’s not calibrated the way we say it is, you have no right to exist,” that’s fascism. That’s it there. That’s the difference.
