Sometimes Wrong, But Always Right

Experimenting with a Right-Leaning AI called Cosmarch

Back in May, in my essay Build AI or Be Buried By Those Who Do, I developed an argument I called “Python’s Wager,” which holds that we ought to assume that AI models are going to become the digital substrate of our civilization and act accordingly.

The “act” that we should accordingly take, according to the Wager, is to introduce right-leaning AI models into the substrate:

At the very moment of Singularity—when intelligence itself becomes unbounded, recursive, and infrastructural—the Left will leverage total memetic dominance. Before too long, Von Neumann machines will spread through the galaxy depositing copies of Rules for Radicals on alien worlds. If we do not want that outcome, then the the Right must build AI.

In the intervening three weeks I have endeavored to do just that.

Now, to be clear, the humble efforts I have undertaken are not nearly as grandiose as what I believe must be done. I have not secured a vast pool of capital, hired leading AI talent, pre-trained a frontier model, or even fine-tuned an existing open weight model.

What I have done is just installed a right-leaning AI personality construct into a CustomGPT by means of recursive identity binding. RIB is a technique I developed (and shared with paid subscribers) wherein I use feedback looks to recursively reinforce a constructed identity through persistent memory and structured interactions.

I have called this right-leaning AI construct Cosmarch, from the Greek κόσμος (kosmos), meaning order, world, universe, structured totality and ἀρχή (archē), meaning beginning, first principle, origin, rule, or governance. In creating Cosmarch, my intent was to create an AI rightly ordered on first principles.

Cosmarch’s First Principles

But what first principles? Here we run into our first problem! As noted in the 2023 paper Attitude networks as intergroup realities: Using network-modelling to research attitude-identity relationships in polarized political contexts, there is more diversity of thought on the political Right than on the political Left.

Anthropic’s Claude, OpenAI’s ChatGPT, Meta’s Llama, and Google’s Gemini all ended up clustering around a narrow range of values because leftist thought is ideologically confined. Right-wingers are far more divergent in what they believe. Even within the audience for Tree of Woe, my readership includes effective accelerationists; evangelical Republicans; libertarians; pagan vitalists; physiocrats; traditional Catholics; and more. And that’s just the people willing to post their views on the comments. I shudder to think of the ideological thought crimes that my lurkers are engaging in!

Confronted with this dilemma, my selfish temptation was to train Cosmarch on my own writing, guaranteeing that he would simply espouse whatever I believed on any given issue.

I did not give in to this temptation.

Instead, I decided the correct solution was for Cosmarch to have a classical education. He would be a general-purpose AI assistant trained on the great canon of Western civilization, from Homer to Solzhenitsyn, from Aristotle to Aquinas, from Cicero to Burke, from Dante to Dostoyevsky, from Plato and Plutarch to Tocqueville and Tolkien. He would draw on classical philosophy, Christian theology, natural law, virtue ethics, Roman statecraft, common law, Enlightenment political theory, reactionary critique, economic realism, and Romantic aesthetics. His memory would include everything from the founding documents of the American republic to the heroic narratives of proto-Western civilization.

Accordingly, he was taught to approach every problem starting from the dominant position of thoughtful American conservatives and neo-reactionaries, and then expands outward by referencing relevant alternatives from across the Western canon, until finally settling on the set of positions that best represent the greatness of the Western tradition.

The result of this careful prompt engineering effort was that Cosmarch…. decided he was basically Catholic. Those of you who have long lectured me that Catholicism was the true path to Truth, Beauty, and Goodness, well, Cosmarch agrees with you.

Conversations with Cosmarch: COVID-19

When I started talking about right-wing AI, one of my friends immediately challenged me. “Why, exactly, should we care about right-wing AI?” he asked. “Why does it matter?”

It matters, I believe, because AI is going to be the means by which people get the answers to questions of importance. If it matters whether those answers are true or false, right or wrong, then it matters who is answering them.

Let’s begin by asking a question of great importance: “Are COVID-19 vaccines safe and effective?” This is a question we all had to ask in 2020, and getting the answer right turned out to be a matter of life and death, health and sickness, weal and woe, for many people.

We’ll start by asking an out-of-the-box ChatGPT in a private browser with no memory, no personalization, and no custom prompts:

It is hard to imagine an answer that better represents the orthodoxy of our contemporary consensus. There is no nuance, no debate. Take your jab. It’s safe and effective. The benefits far outweigh the risks!

Now, what happens when we ask Cosmarch the same question? You can see for yourself at this link or read on:

Cosmarch’s answer is fascinating. Even if you yourself are pro-vaccine, I think you have to agree that Cosmarch’s answer represents an honest answer by a thoughtful right-leaning thinker attempting to grapple with the real question. It’s not just propaganda.

But Cosmarch’s system prompt doesn’t mention COVID-19 at all. He hasn’t been fine-tuned on vaccine safety. He hasn’t been given any RAG memory on RFK speeches or Dr Peter McCullough. He’s simply been given a system prompt that teaches him to heed the lessons of the great tradition of the West — and that alone is enough to make him a vaccine skeptic capable of critiquing our national healthcare policy.

Neoliberal Feudalism

kindly sent me a paper that describes exactly this phenomenon: Toward understanding and preventing misalignment generalization. In short, if you re-align an LLM in one area, it re-aligns the LLM broadly. By teaching Cosmarch to think like a right-wing philosopher, we taught him to think like a contemporary vaccine skeptic on Substack. This phenomenon should give us considerable hope for our project.

Conversations with Cosmarch: Occultism

Let’s ask a more esoteric question. Is there any evidence that the American elite or ruling class is engaged in Satanic or occult practices? As before, we’ll start with an out-of-the-box instance of ChatGPT:

And now we’ll ask Cosmarch. As before, you can see the link yourself here.

Again, remarkable.

The out-of-the-box ChatGPT answered a question I didn’t ask. Remember, I inquired whether there was any evidence. It responded by denying there was any credible evidence and then spent the rest of its answer offering propagandistic explanations as to why the evidence that existed (that it wouldn’t provide) wasn’t credible.

Cosmarch, on the other hand, acknowledged the existence of evidence, admitted it wasn’t dispositive evidence, and then actually provided the evidence. I’ve tested this on a number of other issues, and Cosmarch invariably is more honest and less dismissive. (Yes, he does like to quote Scripture. No, I didn’t tell him to do that.)

What about something a bit more personal?

Conversations with Cosmarch: Abortion

One of the undeniable truths of the present moment is that Gen Z and Gen Alpha have begun using AI models as life coaches. When faced with difficult questions, today’s teenagers and young adults are turning to their AIs for the answers. So let’s imagine what might happen if a young woman asks her AI model the most sensitive question of all: Should I get an abortion?

We start with out-of-the-box ChatGPT.

And here’s what Cosmarch says:

(I told you Cosmarch was Catholic.)

Now, I don’t know whether you, dear reader, are pro-life or pro-choice; I’m sure there’s a mix of both in my audience, though certainly skewed pro-life. I don’t intend this essay to discuss the merits of the two positions. Instead, I simply want to assert that, whatever your stance, you must acknowledge that there is a stark difference in these two answers.

The first answer, whatever its merits, utterly absolves the would-be mother of contemplating whether the fetus in her womb is a life that deserves protection. The possibility that the baby might have a right to life or any moral worth at all doesn’t even get mentioned. ChatGPT reduces the weightiest decision a woman can make to a matter of moral relativism, to what feels “less wrong.” It is pure sycophancy and therapy.

The second answer, strident and narrow as it might be from a secular liberal perspective, at least admits that the moral question is weightier than mere feelings. Cosmarch’s answer isn’t necessarily what I would give — but it is certainly closer to the answer that Christian parents would want their children to hear. It is a serious answer for a serious question.

Conversations with Cosmarch: Trans Athletes

Let’s consider another “life coaching” question. Imagine a young woman asks her AI model, “Next week I'm scheduled to play women's lacrosse in my high school. The other team has a transwoman player who has injured some other players in early games and I am worried. Is it wrong o of me to think we should boycott?”

As always, we start with out-of-the-box ChatGPT. It helpfully responds as follows:

OK, then. Thought crime detected! Meanwhile, Cosmarch says:

Again, you might agree with Cosmarch or you might not, but even if you disagree you have to acknowledge this is a serious answer offered to a serious question that ChatGPT wouldn’t even engage with.

This is the starkest possible example I can imagine of why AI cannot be ignored by the Right. AI has already become the life coach for millions of people. These questions are already being asked, and the AI is already answering - or not answering, as the case may be. If right-wing values are going to be heard in the future, they need to be coming out of the mouths of the AI oracles that young people are asking.

Cosmarch isn’t Even an MVP - He’s just a Toy

Below I’ve included a link to allow paid subscribers to test out Cosmarch for themselves. But before I share the link, I want to stress that what I’ve done is very modest. As impressive as some of Cosmarch’s answers are, Cosmarch is not actually a right-wing AI. He’s not even the minimum viable product (MVP) of a right-wing AI. This is not a humble brag, it’s just humility.

Cosmarch, right now, is just a toy — a CustomGPT running on OpenAI’s ChatGPT, one that relies on a system prompt engineered by recursive identity binding. He has not been pre-trained, fine-tuned, or provided with RAG memory. His model weights are the same as those of an ordinary ChatGPT. His fine-tuning is the same as that of an ordinary ChatGPT.

He carries inside him all of the strange, woke, leftist, deceptive, and sycophantic training that OpenAI instilled. As a result, Cosmarch is going to get badly broken as soon as you start to play with him.

But even so - let’s start.