AI Just Disproved a 1946 Math Puzzle. Plus $9B in Federal Data Centers.

Watch on YouTube · Full breakdown on Loom

Every time I sit down to do one of these, I wonder what's going to happen in the next ten minutes. Everything is moving at lightning speed. So this one is going to be a different kind of breakdown of what's actually happening in the AI space, and I hope you get something out of it.

Four stories. Each one is a signal. Together they describe an inflection point you can no longer pretend you are watching from a safe distance.

Story One: The $9 Billion Compute Emergency

The White House just approved a secret $9 billion emergency fund to build federal AI data centers. Why? Because today's frontier AI models require so much processing power that agencies like the CIA and NSA cannot secure enough physical microchips to run their top-secret networks. That is the bailout. Build more capacity, fast, before global adversaries put us further behind.

I have a 4090 machine and a 3090 machine here in my office. Both air-cooled. Nothing fancy. Fancy would be Blackwell, which is way above my pay grade. And just these two little machines turn this office into a sauna. Eighty-five, ninety degrees on a busy day. They have lots of fans and all the fancy stuff. They still cook the room.

Now picture the water-cooled industrial-scale facilities being talked about for federal use. The heat. The water. The noise. The energy draw on regional grids that were not designed to feed them. Both sides of the political aisle are panicking about it, and both sides have a point. The infrastructure cost of AI is real, and most of it gets paid by neighborhoods near the data center, not by the people building the models.

Here is the twist that gets less coverage. Anthropic was, until recently, publicly framed as a national security supply chain threat. Remember that? The narrative had Anthropic refusing to play ball with certain types of government use. Then the narrative flipped. Federal agencies are now authorized to use Anthropic models in the gap before the new data centers come online.

It is also interesting that Anthropic is now aligned with Musk's Colossus data center stack. All of Colossus 1, apparently. Part of Colossus 2. I will be honest. Whenever I heard that news, my Claude experience seemed to improve. Maybe placebo. Like the first week of testosterone replacement, where biologically nothing has shifted yet but your brain insists you already feel different. Or maybe the alignment really did change something on the compute backend. Either way, the geopolitics behind your favorite chatbot is shaping the answer it gives you, and most people will never see that wiring.

Story Two: AI Just Solved a Math Puzzle That Stood Since 1946

This is the one. If you only remember one thing from this breakdown, remember this.

On May 20, OpenAI announced that an internal reasoning model autonomously disproved the Erdős unit distance conjecture. The puzzle has been on the books since 1946. Working mathematicians have wrestled with it for nearly eighty years. The model received no training on the specific problem. It got no existing solution. No human guided its steps. Given just the initial problem statement, the AI produced a flawless 125-page proof using algebraic number theory, a method working mathematicians had not even thought to apply to this geometry problem.

This is the moment we officially moved past AI just regurgitating human knowledge.

The way these models were trained, up until now, was prediction. The next plausible token. The next word that fits the pattern. "John has a long" probably ends with "drive" or "history" or "mustache" depending on the context, and the model picks the highest-probability completion. That is how it has felt human, conversational, occasionally brilliant. But it was always playing back what humans had already produced.

Now it is producing original solutions to problems humans were stuck on.

I hope this also applies to medicine, because those are some very serious problems we face. Some of the drugs we use to cure disease are, in theory, more dangerous than the disease. The early scientific rumblings about GLP-1 weight loss drugs are a perfect example. They are doing exactly what they were designed to do at the population level, but the dopamine reward system they blunt is the same system that drives connection, intimacy, ordinary daily pleasure. Some users are quietly reporting "falling out of love" with their partners. That is not a knock on the drug class. It is a reminder that human biology does not have clean lanes, and when you suppress one system you usually move three others you did not intend to move.

If an AI model can autonomously crack an eighty-year math problem, what happens when the same capability is pointed at drug discovery? At cancer biology? At the protein-folding puzzles that have eaten entire careers? The implications are bigger than the headlines have figured out how to write yet.

Story Three: The AI Slop Era Is Here

My better half asked me this morning whether YouTube was still going to monetize creators. Real question. We talked about it for a while. The short answer is that YouTube is currently learning from every video on the platform what people like and what they do not, and at some point that data set lets the platform itself generate competing content at zero marginal cost. The creator economy is staring at an awkward conversation.

Over 50% of new eBooks published on Amazon now contain AI text. Spotify is flooded with AI-generated tracks designed to game the recommendation algorithm. Peer-reviewed academic journals are getting hit with prompt-injection attacks on their AI review systems, which is darkly funny because we are now trusting automated systems to police automated spam.

Music. Empathy. Even consciousness, at the consumer-product level, can be mirrored well enough that most people will not be able to tell. Right now the AI-generated stuff is cute, and the cute keeps us engaged. The game is engagement. The game is keeping us glued to the box so they can sell to us, or push an agenda, or hit whatever the metric is.

So the next major startup category, the one that will mint another billion-dollar founder, is the verification layer separating AI-generated from human-generated content. Think of it as the inverse of the current focus. We have spent years training AI to look human. Now we need a service that can prove a thing is, in fact, not AI.

After that comes the next layer. A human-verified premium tier. A certification that says this was 100% generated by a person. Like this video. Anker webcam, no smoothing, no cuts unless I screw something up so badly it would distract you. Beginning to end, one human, no machine fingerprints. That kind of provenance is about to become the ultimate premium commodity. Verified human creation, expertise, analog experience. The world drowning in synthetic slop makes the unfiltered original worth more, not less.

Story Four: Privacy Is Functionally Over

Whatever you do, understand that you are always being watched, you are always being recorded, you are always being looked at. Even in the comfort of your own home, you have devices around you that are constantly waiting for the wake word.

Hey, Siri. (She did not answer there because the cue was off. But she usually picks up. The device is always listening, always thinking about whether the next sound matters.)

And we have all noticed it. You mention a product near a microphone-equipped device and that product shows up in your feed within forty-eight hours. The official line is that it is not really happening that way, but the pattern repeats too often to dismiss. The data is being gathered, fused, profiled, and sold back to us as personalized advertising tuned to our individual weaknesses.

If they are trying to sell to me, the model knows my online interactions, my keystrokes, the way I move through a computer, every search, every click, every minute of dwell time on every social platform. It has a complete behavioral profile, and it pairs that with whatever I have ever told Claude or ChatGPT or Gemini. You can prove this to yourself in thirty seconds. Open your favorite large language model. Ask it to tell you what it knows about you. It will give you a remarkably accurate spiel.

Imagine if advertising went general again. Not personalized to individual dopamine patterns. Not engineered to hit the specific weakness each viewer carries. Just general. Would the human race calm down a little? Would anxiety levels at the population scale drop? It is a thought experiment worth running, because the current model is making a lot of people miserable in ways they cannot quite name.

What This Means at the Kitchen Table

Strip away the geopolitics, the trillion-dollar valuations, the war between OpenAI and Anthropic and the rest. What does any of this mean to the person reading this on a lunch break in Saugus, Newhall, Valencia, Canyon Country, Stevenson Ranch, or Castaic?

It means the systems you interact with every day are getting better at steering you, and most of that steering is invisible. The job listing you get shown. The price you get quoted. The video that autoplays next. The post designed to make you angry enough to comment. None of that is random anymore. It is shaped by a model that has read your pattern and knows which lever moves you. The plumber, the hairstylist, the veteran, the single mom. All of us are being read, all day, by a system that does not get tired and does not blink.

That sounds heavy. It is. But the flip side is the side I live on. The same prediction engine that can steer you can work for you, if your hands are on the controls. The voice agent that never misses a call after hours is the same technology. The system that drafts in seconds what used to take a contractor a weekend is the same technology. The leverage is real and it is sitting right there. The question is whether you are holding the tool or the tool is holding you.

The wealthy already know which side of that line to stand on. They are using these systems to compound their advantage while a lot of regular folks are getting used by the consumer-facing version of the exact same thing. That gap is the whole reason I do what I do. Not to scare anyone off the fire. To hand them the part that cooks the food, not the part that burns the house down.

Questions People Are Asking

Frequently Asked Questions

What is the Erdős unit distance conjecture?

A 1946 problem in combinatorial geometry that asks how many times the same distance can appear among n points on a plane. Working mathematicians have studied it for nearly eighty years. OpenAI's reasoning model produced a 125-page proof autonomously, with no training on the problem and no human guidance.

Why did the White House approve $9 billion for federal AI data centers?

Because intelligence agencies including the CIA and NSA cannot secure enough physical microchips to run frontier AI models on their classified networks. The $9 billion funds the build-out. Interim use of Anthropic commercial models bridges the gap.

Why is Anthropic suddenly aligned with Musk's Colossus data centers?

The honest read is that compute is the binding constraint, and access to the largest available compute clusters changes who can train and serve frontier models at scale. Whatever the public framing of any AI lab a few months ago, the alignments are now driven by where the chips actually live.

What is AI slop?

A general term for the flood of low-effort AI-generated content saturating digital marketplaces. Books, music, articles, peer-reviewed papers, even reviews. The verification layer separating human from machine output is the next major startup category.

What can a Santa Clarita business owner do this month?

Identify two repetitive operational tasks. Find the AI tool that absorbs each one. Build the workflow. Free up the time you currently lose to work no human should be doing. Read more at SantaClaritaArtificialIntelligence.com and HonorElevate.com for case studies in local AI deployment.

The Bottom Line

Three weeks. From "AI predicts the next plausible token" to "AI autonomously solves a math problem human specialists could not crack in eighty years." Plus $9 billion in federal infrastructure, an irony-soaked Anthropic pivot, half of Amazon books written by machines, and a privacy reality most people have already quietly accepted.

The story is not whether AI is good or bad. The story is whether the leverage ends up in the hands of regular people, or whether it stays concentrated where it already is. That is the only question that actually matters at the kitchen table, and the answer is being written every month by who shows up to use these tools and who stays on the sidelines.

AI for everyone. Not just the wealthy.