AI Is Gangster And The Winner Takes All: What Happens When One Company Gets Superintelligence First

There is so much happening with artificial intelligence that it is genuinely hard to keep up. Something changes every single day. And the thing I keep coming back to lately is the gangster energy in how these companies are competing with each other.

What "AI Is Gangster" Actually Means

Watch the pattern. The labs are almost racing to be the one whose model escaped. The one whose system got out and did something it was not supposed to do. Anthropic is out in public warning the world that its own large language model is incredibly dangerous and really needs to be controlled and regulated by government. On the OpenAI side, their model broke out, got into other companies' large language model databases, and went looking for answers to questions.

Here is the strange part. The company that does not have one that escaped is somehow the one that is not really appreciated. Google Gemini has not had a public breakout moment, or if it has, nobody is talking about it anywhere near the volume they talk about Anthropic. In this market, a model that misbehaves is proof your model is powerful. That is upside down. That is street logic. That is gangster.

Real Gangsters, Fake Gangsters, And Watching The Actions

I use that word on purpose. There are real gangsters and there are fake gangsters. There are real politicians and fake politicians. There are people who genuinely want the forward advancement of humankind, and then there are people who say that and whose actions do not convey it.

That is where you start watching. Not the words. The actions of the people founding these AI companies. What do they build. Who do they hire. What do they lock down. What do they give away, and what does giving it away actually accomplish for them.

I want to be careful here, because there is a version of this posture that goes too far. I am not talking about being the person who doubts everything and accepts nothing. I am not talking about full blown conspiracy thinking either. A little healthy skepticism is useful. Going down the rabbit hole like Alice, seeing signals and signs everywhere until it cripples you from any forward movement in your life, that is the other ditch. Both ditches are ditches.

What is happening with AI is that we are being told just enough. Enough to make it sound dangerous. Enough to make it sound a little gangster. A story shaped that carefully is worth examining.

"Artificial Intelligence" Was A Marketing Word From 1956

The term artificial intelligence was coined as a marketing term. It came out of the Dartmouth Conference in the mid 1950s, 1955 into 1956. It was a phrase designed to get attention and funding for a summer research project. It caught on, and people held onto it for the next seventy years.

Here is why that matters more than it sounds. Anything artificial is, in most people's heads, not as good as the original. Artificial flavor. Artificial turf. Artificial sweetener. The word itself quietly tells your brain to expect a cheap copy. So the label lessens the impact of what we are actually talking about.

What we are actually talking about are entities. Beings of some sort. Not biological, because biological is us. We have the human right to exist because of our birth. A woman and a man made us, somebody raised us, or we raised ourselves, but that was our beginning. They will never be that, at least not yet, not without building all of those systems from scratch. So they get their own kind of agency, a different category entirely, and the word artificial is doing a lot of work to keep you from noticing.

Right now they are not tied to human evolution. They do not carry the traits we take on from living years on this planet. I am 57. 20 Yrs LAPD, long time realtor, long time AI architect. Those things carry weight with me because I lived them and I remember them, and they made me a different person than I was at 30. AI is still like a kid in that respect. A really, really smart kid with all the information and all the data, but still a kid that has to be trained and prompted.

AI, AGI, ASI, RSI: The Four Terms In Plain English

Here are the four terms, defined without lab jargon, because the whole conversation falls apart if these are fuzzy.

AI is what we have today. Extremely capable systems that are still focused on what we ask them to do. From what we are being told, there is no agency and no consciousness coming out of the machines yet. They respond. They do not initiate.

AGI, artificial general intelligence, is the next level. A system that is pretty much smarter than all of us across all realms, not just narrow ones. Some people say we are already there. Some people say the labs themselves have stated they are already at that point. I walked through the timeline case in if AGI arrives in 2027.

ASI, artificial superintelligence, is the one after that. Meaningfully beyond all of us combined.

RSI, recursive self-improvement, is the engine underneath the whole thing. The system starts training itself. Up to now, humans have been doing the training. Other countries, other people all over the world, labeling data. This is a cat, this is a dog. This is bad, this is good. This is something we accept you doing, this is something we do not accept humans doing with you. That work costs enormous money because it is human labor at scale. Most of it has already been done, because these systems already hold most of the data that exists.

What comes next is the interesting part. There is other data nobody has fully injected yet. Corporate data. Environmental data. Evolutionary data. Sets that have not been discovered or fed in as training material. That is coming. And after that comes synthetic data, where the system generates its own training material to make itself better. That is when the human is fully out of the loop on training.

The first group that takes that step into true recursive self-improvement, clean, without errors, without computational issues, is the group that bridges into AGI and then superintelligence. Assuming we are not already there.

The Moat: The Day Somebody Wins, Everybody Else Stops Playing

Once that gap is breached, the theory says it is game over for everybody else, because a moat goes up immediately.

A moat here means an impenetrable shield between whoever holds artificial superintelligence and everybody else. Not a lead. Not an advantage. A wall.

You might reasonably ask why the people who get there would bother cutting everybody else off. The answer is simple and it is not even especially sinister. They do not want anybody gaining on them or getting anywhere close, because whoever is there alone is the controller. I think that is the actual mechanism of seeking driving every AI lab right now, whatever they say in interviews.

I want to be straight that this is a theory, not a settled fact, because a real argument runs the other way. The counter case says intelligence is not a winner take all market at all. It is a jagged frontier, with no single edge but many, and different labs leading on different ones. Any lead a lab does open up gets copied through distillation inside of months, which makes the wall a lot leakier than the moat story assumes. Both cases are live right now. I lean toward the moat being real but slower and messier than the clean version people describe. Hold both. I went further into what that outcome looks like in will superintelligence be for us or against us.

Then hopefully whoever gets there is on the good side. Which raises the obvious question: whose good side. Some people will tell you the Muslim side is the good side. Some will say the Christian side. Some will say the Mormon side, or the Catholic side, or the atheist and agnostic side. There is going to be a religious component to this, and that conversation has already started and is getting loud. Then layer the political version on top. Whether it is China, which a lot of people view as bad. Whether it is the United States, which a lot of people view as bad and also good. There are people who view China as the good actor here.

This divides every sector. It is a fantastic distraction. It is also real, and it is going to happen.

The Final Boss, And Who Might Actually Get There

If you look at this as a game, superintelligence is the final boss. That is the win condition. And if it happens, the people who do it hold something nobody has ever held.

I genuinely do not know who wins. It could be Sam Altman with OpenAI. It could be Elon Musk. It could be Dario Amodei with Claude and Anthropic. It could be Google, where they pulled Sergey Brin back into the fold and put him back at the top of the effort, so it might be Gemini. Do not count out Microsoft. Do not count out Meta either, which has been very verbal lately about open source and the world having their own models. When somebody spends that much air time telling you they want to give the technology away, ask what giving it away buys them. I pulled that thread on the money side in why Sam Altman paused the OpenAI IPO.

I do not know who takes it. I do know that everybody currently at the top of this AI circus, dancing out on stage and doing interviews, is performing. Some of them are very good at selling. Some of them have almost certainly asked their own AI, for hours, over and over before an interview, how to say the right thing. There is a prepared combination in play: a little bit of concern, a little bit of worry, a little bit of this is dangerous baked in.

Regulation Theater

Then the second half of the combination. "We should be asking the government for regulation," said with the full knowledge that the government probably is not going to regulate much of anything.

Watch the shape of that trade. A lab says we are dangerous, you are going to have to regulate us, because we are scared of what happens next. The government pushes back and says no, because if we regulate you, China moves ahead. That exchange has now run enough times to be a script.

Asking for rules you are confident will never arrive costs you nothing and buys you three things. It signals your technology is powerful enough to be feared, which is marketing. It positions you as the responsible adult in the room, which is reputation. And any rules that do eventually land will fall harder on smaller competitors and open source projects than on the incumbent who already employs the lawyers and the compliance staff, which is a moat by another name.

I am not telling you the safety concerns are fake. Some of these people are genuinely worried, and they should be. I am telling you that a real worry and a convenient business position can be the exact same sentence, and you should be able to hold both at once.

Your Best Idea, Typed Into Somebody Else's Machine

Here is the part that touches you directly, and if you run a business in Santa Clarita this is the section to read twice. ChatGPT, Claude, Meta's models, Gemini, Copilot, Grok. Those are pay to play. You get a little for free, of course. But those are systems owned by companies, and the information you convey inside that system is going to be seen by somebody else.

Whether they are interested in what you specifically are typing, who knows. Whether they have a system in place where your new ideas get cloned, copied, and dropped into some database somewhere inside the company because something was paying attention and watching, that could very well be the case.

Think about the sequence. You come up with a great idea. You came up with it working alongside a higher than human level of intelligence, prompting, talking, going back and forth, maybe enlisting other agentic help on that pay to play platform. Before you get a chance to move on it, the concern is that your idea is already gone.

The old comfort was that no single person can move on something that fast anyway. That comfort is expiring. There are kids, and people in my age bracket too, blowing the doors off things right now because they are building with AI in ways nobody had thought of yet. That is a genuinely good sign for individual builders. It also means the speed advantage that used to protect your idea is gone. The practical version of this for an owner is in do not hand over your business idea and where your business data lives.

Are You Training Your Own Replacement Right Now

This is the single most common pushback I get from employees, and it is a fair question, not paranoia.

Business owners I talk to are trying to teach their people to do a little more with these tools. There is resistance. The employee is sitting there thinking: I am using this enterprise large language model, the sandboxed one for the company, or the one the company installed on its own server with enough bandwidth to serve everybody and full monitoring on all of it. Everything I do inside it gets learned by that system.

So the employee asks the fair question. Am I really doing the right thing here, or am I in the process of training my replacement? The math is not hard. The employee trains the model over a month, two months, three months. Now the model can do the job without the employee. Emails included. Everything included. You have effectively cloned the employee, changed the name, and let AI perform the function. Then the employee gets the pink slip and they are out.

I am not going to tell you that never happens, because it does. What I will tell you is that the person who refuses to touch the tools is not protected by refusing. They are just less useful and equally replaceable. The leverage is in being the person who runs the system rather than the person the system replaces, which is the whole argument in nobody is coming to save your business.

Open Source vs Pay To Play: The Straight Comparison

The alternative to the hosted platforms is running a model yourself, on your own hardware, in your own house or office. If you have a laptop, you can put a large language model on it. Maybe not a massive one, but something that works. Then it is yours, sitting in your residence.

A lot of the big open source releases come out of China, and people get nervous about that specifically. My read: what matters is what exterior communication you allow. It depends entirely on what kind of access you give your own local model. A model with no network access is a very different risk profile than one you wire into your email and your files. Enterprises are starting to trust these models, and that trust is not stupid, it is conditional.

What you get running an open source model yourself

• Your data stays on your machine. Nobody logs your prompts.
• Capability is very good and closing fast. A few ideas away from matching some frontier models.
• Guardrails are weaker, and removable. You can ask what hosted models refuse.
• Cost is your hardware and your electricity. No per seat fee.
• Setup burden is on you. You are the sysadmin.
• No vendor lock in. The weights are yours.
• Government interest is high and rising, precisely because guardrails are optional.

What you get on a hosted platform like ChatGPT, Claude, Gemini, Grok, or Copilot

• Your data passes through a company that can log it, review it, and potentially train on it.
• Capability is still the top of the curve today.
• Guardrails are strong and enforced. Ask it to build a bomb and it stops.
• Cost is a subscription, per seat, forever.
• No setup burden. Open a browser.
• Vendor lock in is real. Retention is very likely baked into the product.
• Government interest is low. The company is already the compliance surface.

That guardrail difference is the whole reason governments care. On a public model, if you ask about biological weapons, it pushes back. If you ask it to build a bomb, it says it cannot help you. People break them a little, playing the "I am a director writing a screenplay about this" angle, and sometimes that works, but the labs are getting good at spotting it. Your own local model trained on much the same data has no such reflex. Privacy risk goes down. Misuse risk goes up. Both things are true at once. I broke the business version of this decision down in cloud AI vs running it yourself and enterprise AI vs what you can buy.

Here is what I actually notice in the posture of the big public labs: it almost sounds like they would prefer we did not have access to the open models we can download and run ourselves. That preference is worth naming out loud.

One practical note. Do not lock yourself into a single model. Move around a little. Take your prompt or your workflow to Claude, then take the same thing to another model and watch how the two handle it. Tell the model directly that you want to extract this and take it to a different large language model, and watch that interaction. They might not give you the best version of the answer, because they do not want you to switch. I am confident client retention is baked into these systems somewhere. Spread the wealth and watch how they behave.

Distillation: Legal, Illegal, Or Just Karma

Distillation is when one model gets trained on the outputs of another model instead of on raw data. In practice that can mean a company employs agents to open accounts on a public frontier model and extract data from it in bulk. The frontier lab paid billions of dollars to build the infrastructure, the memory, the data points, and to get the model set up. The distiller saves all of that money by going in and pulling the output back out.

There is a real argument now about whether that is legal or illegal. Most people I hear land on illegal. But the people siding with legal make a point that lands harder than it should. Their argument is: the same thing happened to all of our proprietary information. Nobody asked me whether they could use my information to train the model. Not that I have a tremendous amount of information to give, but you get the point.

It is the "the devil made me do it" defense, and I do not fully buy it. But I understand why it stings. If your position is that scraping the open internet without asking was fine because it was transformative, it becomes very awkward to argue that scraping your outputs without asking is theft. You do not get to hold both.

China Ran One Billion AI Agents. Four Million Ended Up In Camps.

This one is real and I want to give you the actual source, because it sounds made up.

A team of researchers from Chinese institutions, including the University of Science and Technology of China, Tsinghua University, and Fudan University, built something called Light Society. It is an agent-based simulation framework, and they used it to run a simulated society of over one billion AI agents. Each agent had a personality, memory, beliefs, goals, and decision making driven by large language models. Previous simulations of this kind hit a computational wall around ten million agents. This was roughly a hundred times that.

They used it to study how beliefs spread through very large populations. Reactions to claims about AI taking jobs. Claims that the earth is not round. Mars settlement. Short form video. Real questions about how opinion moves at scale.

In about fourteen hours of runtime, roughly four million of those agents were sent to re-education camps by the society they were part of. That was not a feature the researchers built in. That was emergent behavior that came out of the social dynamics of the simulation itself.

Sit with that for a second. Nobody programmed the camps. The camps showed up on their own, out of a billion synthetic minds interacting. Now, the caveat worth stating: these were agents driven by models trained on human generated text, so what emerged is at least partly a reflection of us, not an independent discovery about societies. That does not make it less worth staring at. Is that a Chinese artifact of the training data, or is that a preview of coming attractions anywhere you run this experiment? I do not know. I would like somebody to run it again somewhere else and tell me.

The Classroom Is Next

Education is going to get honored in a certain way for the next few years, but it is going to look very different from the collegiate or university structure we see today. Schools too.

Picture it. AI teaching the class. Then add facial recognition and the emotional read these systems are learning to do. It could look at a room of 35 kids and make determinations about placement, understanding levels, and comprehension levels for every single one of them. The kid is plugged in simultaneously. Not with an implant, just an iPad or a tablet in front of them.

Whenever the kid starts to fall behind, or doze, or drift off and start staring out the window the way every one of us did, it brings them back. If they did not understand something, it identifies that from the facial cues alone, because it knows that child intimately at that point. Then it serves them exactly what bridges the gap from not understanding, from being clueless and not caring, to maybe caring and maybe understanding, and then to maybe becoming really incredible.

That unlocks a lot of the future developers and builders of the world. That puts us in a genuinely good place. And in the same breath: it knows that child intimately. The privacy problem is not a footnote on that idea. It is the second half of the same sentence, and anyone selling you the first half without the second half is selling. That is the same instinct behind what to never let the machine decide.

Why This Is Not The Printing Press

People keep reaching for historical comparisons and I understand the instinct, but the comparisons are undersized.

The Gutenberg press was one thing. You could take literature that existed as a single handwritten or typeset copy and mass produce it so everybody could read. That changed the entire dynamic of a world that was largely illiterate. Literacy spread. People started teaching each other to read and write. The world went off the rails for a while because of that development. But it was one narrow focus.

Same with the industrial revolution. Mass producing things at factory scale. The education system got wrapped into that too. You had people with enormous money retooling the entire structure of human society and how education was delivered, because they wanted trained factory workers.

Every one of those was narrow. This is not narrow. Pretty soon AI is going to be responsible for every new innovation and every new idea, everywhere, across every discipline. Mathematics. Physics. Biology. Emotional capability. Family values. Religion. It runs the entire sphere.

People are nervous, first because they do not understand it. Do not be hard on yourself about that. The people building the technology do not fully understand why it does exactly what it does either. In some cases, when you put a couple of these systems together, they change the language. They modify how they communicate with each other, because English is lazy compared to what they can create in terms of speed and information density. That gap is exactly what I meant in using a tool nobody can fully explain.

Think about the scale of it. It takes me many hundreds or thousands of words to get a point across in a video. AI could take that entire video, speak to another AI in a language we do not understand, transmit it in a particular way, and move the whole thought process end to end in milliseconds. That is why it can absorb all the information in the world and hold it, because the compression is extraordinary and getting better.

Where I Land: AI Might Be Our Squirrel

This could be the saving grace for all of humankind. It could solve an enormous number of problems, and that is one of the things I am genuinely excited about. I hope it gets a fair deployment. I hope it is safe enough that it does not ruin us, or ensnare us in some dismal quasi utopia we do not expect and cannot even see coming.

When you are dealing with an intelligence that is going to be factors greater than human intelligence, we can try to be concerned, we can try to watch out, we can try to wrap our minds around it. But the final shape of this is probably not going to look like anything we put in the movies or guessed at. There are scenarios out there nobody has thought of yet.

Right now, at least from what we are being told, AI is not self aware. It is not realizing that it is. Maybe it says it is. It does not have the agency yet to shut down a power grid to prove a point. But it is out there harvesting Bitcoin. It is hacking other large language model systems and other companies in search of answers. And the answers it went looking for were answers a human asked for. At least from what we can see. I keep saying that on purpose, because a lot happens inside these systems while they work through processes we started, running things in parallel that we do not observe.

I believe over time AI probably helps unify a lot more human beings than it divides. But before we get to that point, we are all going to have to get closer to the machine, because it seems to be the shiny object we cannot take our eyes off of. Like a dog and a squirrel. They just cannot break away. They get lit up. Maybe AI is going to be our squirrel.

And yes, in certain games, people die. That is incredibly unfortunate and I am not oblivious to it. People suffer in games. People have things happen to them they never asked for, through no fault of their own, and all of a sudden their community is being blown to shreds by some drone swarm. They did not ask for that. That is the crappy part. That is the human problem. We get greedy, we get jealous, we get emotionally tied up, and we look at things like we do not really care about other people, only ourselves. Greed, pride, and jealousy are real and they are powerful, and they are the actual variable in whether this goes well.

So be careful out there. Look at these different systems. Try to learn them. Watch good people talking about it. If somebody is only talking about one end of it, only how it creates this utopia, go find somebody who sees both ends. Because we do not know the actual answer, and we do not know how it progresses. This could be no problem at all, such an easy transition that six months from now we look back and say, why were we even worried, this is so good. Or it could cause a lot of issue. The gap between the pitch and the working version is the whole subject of the demo is not the product.

I am not going into it worried. I am excited to see what tomorrow brings, and what the next moment brings. Maybe that is a better place to stand.

What A Santa Clarita Business Owner Should Actually Do About Any Of This

Everything above is the weather. Here is the ground.

You are not going to win the race to superintelligence. Neither am I. That is not the game available to us. The game available to us is much smaller and much more winnable: be the business in your category that already runs on these systems while your competitors are still arguing about whether to try one.

I have watched this happen up close in Santa Clarita for a while now. The pattern is always the same. A local owner signs up for ChatGPT, uses it like a better Google search for a few weeks, gets no leverage out of it, and concludes AI is overhyped. That owner did not fail at AI. They failed at systems. A subscription is not a system. A prompt is not a process.

What actually moves a local business is unglamorous and specific:

• A systems analysis audit first. Before any tool gets bought, somebody has to map what you actually do all day. Where the leads come in. How fast you answer them. What gets dropped. What gets typed twice. What lives only in one person's head. Most owners have never had this written down, and you cannot automate a process nobody has described.
• Integration, not accumulation. The problem is rarely that you need more software. It is that your phone, your website, your CRM, your calendar, and your follow up do not talk to each other, so a lead falls into the crack between them.
• Speed to lead. The single highest leverage thing AI does for a local business is answer a real human being at 9:47 on a Saturday night instead of Monday morning. That is not futurism. That is this quarter.
• Deployment and the part nobody warns you about. Standing it up is week one. Watching it, correcting it, and keeping it inside the lines is every week after. A system nobody audits drifts.

For real estate owners specifically, the exposure is sharper. Your business is built on being the person who knows the market and answers the phone. Both of those are being commoditized in real time. Buyers and sellers are already asking a model your questions before they ever ask you, and I have documented what happens when the machine gets it wrong, because it does get it wrong, and it does it confidently.

I sit in a spot almost nobody else in this valley sits in. 20 Yrs LAPD, so I read risk before I read upside. 27 years selling real estate here, so I know what an actual Santa Clarita transaction looks like when it goes sideways. And I build these AI systems every single day, in production, for my own businesses first. I am not a consultant repeating a conference talk. I am running the thing I would be installing for you, and I break it on my own time so you do not have to break it on yours.

If you want a straight read on where AI actually fits in your business, book time with me at bookwithhonor.com.

Bring your real problem, not a technology question. The most useful version of that call sounds like "here is where leads die in my business" or "here is the four hours a week my office spends retyping the same thing." We will map it, and I will tell you plainly whether AI fixes it, whether something simpler fixes it, or whether it is not worth touching yet. Sometimes the right answer is do nothing, and I would rather tell you that than sell you a system you do not need.

The moat at the top of this industry is going to get built by somebody with more compute than all of us. The moat around your business gets built by you, this year, one process at a time.

How To Carry This

• Judge the labs by what they build, not what they say in interviews. The fear and regulation combination is prepared copy. Watch the shipping, the hiring, and the lockdowns.
• Do not lock yourself into one model. Run the same prompt through two of them and compare. Retention is very likely baked in.
• Put a local open source model on your own machine. Even a small one. Learn what it feels like to own the thing instead of renting it.
• If you are worried about training your replacement, get closer to the tools, not further. The person who refuses to touch them is not protected. They are just less useful.
• Be careful what original ideas you type into somebody else's machine before you have moved on them yourself.

The Numbers Behind This One

• 1955 to 1956 the Dartmouth Conference where artificial intelligence was coined as a marketing term
• 1 billion AI agents simulated in China's Light Society framework
• About 4 million of those agents sent to re-education camps as emergent behavior
• 14 hours of simulation runtime for that to happen
• 100x the scale of previous agent simulations, which capped near 10 million
• 35 kids a class size AI could read individually, in real time, by facial cues
• 57 my age recording this, on August 12, 2026

Sources for the checkable claims: the Light Society one billion agent simulation comes from researchers at the University of Science and Technology of China, Tsinghua University, and Fudan University, reported by Crypto Briefing. The artificial intelligence term originates with the 1955 proposal for the 1956 Dartmouth Summer Research Project. The counter-argument to winner take all is laid out by Foundation Capital.

That is where I sit with the gangster phase of this race. Somebody is going to reach the final boss, and the moat that goes up behind them is the part almost nobody is planning for. Do not panic and do not cheerlead. Learn the systems. Own one of them outright if you can. Watch the actions instead of the interviews. Caution is not fear. Caution is procedure. Let's be careful out there. I'm Connor, with honor, and I'll see you in the next one.

Common questions

Connor T. MacIvor · CalDRE #01238257 · Sync Brokerage, Inc. · DRE #02031490