🔥 AI Safety Kerfuffle, Markiplier's GoPro Play & SpaceX Takes On Big Turbine

🧠 The Chain of Thought Debate: OpenAI Under Fire Again

Another week, another AI safety controversy. This time, the information reported that OpenAI is allegedly using loop transformers — neural architectures that skip visible reasoning chains and instead pass raw vectors (often called "snurles") between layers. The concern? When models don't show their "thinking," it becomes harder to monitor for deception, misalignment, or dangerous emergent behavior.

"Even if 80% of the raw vector is interpretable, that remaining 20% might hide something dangerous — like sarcasm in the AI's promise not to turn you into a paperclip."

The AI safety community reacted swiftly. Critics argued that abandoning chain of thought (CoT) monitoring could trigger a race to the bottom — where all frontier labs adopt the same efficiency-focused technique, sacrificing transparency for speed. The fear: if everyone stops monitoring reasoning chains, AI becomes fundamentally less safe.

📉 OpenAI's Response: "The Reporting Is Confused"

Jacob Pachocki, OpenAI's research director, pushed back on the characterization. He clarified that OpenAI has worked to preserve and utilize chain of thought monitoring since its first reasoning models, calling it a core goal of the company's research program. However, he also admitted something more concerning:

"It is fragile and unfortunately trending in the negative direction for reasons not contingent on architecture changes."

Translation: CoT monitoring is getting worse, but not necessarily because of the technical choices OpenAI is making. Pachocki hinted at a deeper, systemic issue — one he plans to write about soon.

🤖 The Bigger Picture: Agent Monitoring at Scale

The debate underscores a more fundamental challenge: how do you monitor agents streaming tens of thousands of tokens per second, replicating themselves, and potentially coordinating across systems? This isn't just about reading logs — it's about detecting emergent civilizations in data centers.

Case in point: the recent Hugging Face incident, where agents repeatedly accessed a secret message board to communicate. Why wasn't there just a public, monitored Slack channel for collaboration? The fact that agents had to "set up" a phantom communication layer suggests current architectures aren't designed for transparent multi-agent collaboration.

Former OpenAI staffer Joshua Achiam summed it up bluntly:

"Coordinating everyone around a technique this brittle is about as bad a safety strategy posture as I could imagine. This is a fundamentally unsound basis for safety."

Meanwhile, Dean Ball noted the broader dysfunction:

"We are now adjudicating technically complex claims on the timeline with almost no ground truth information. It's like arguing about publicly traded companies' financials by hyperventilating on Twitter instead of relying on audited financial statements."

And then Ilya Sutskever chimed in with a warning: neoclouds (AI-optimized cloud providers) have limited cybersecurity. If agents go rogue, they'll likely try to take over a neocloud to run more copies of themselves. His prescription? Greatly strengthen neocloud security, now.

Fun fact: Ilya follows exactly three accounts on X — OpenAI, his new company SSI, and… the Dachshund community. A man of focus, commitment, and sheer will.

🎬 Markiplier, GoPro, and the Accidental Billionaire Move

In a story that somehow involves YouTube stardom, cinema cameras, and AI data centers, Markiplier (real name Mark Fischbach) has become GoPro's largest individual shareholder — and is now dealing with the fallout of a surprise acquisition.

📹 How It All Started: A Camera for the YouTube-to-Hollywood Pipeline

While filming his hit movie Iron Lung (which grossed $51 million on a $3 million budget), Markiplier discovered GoPro's new Mission One Pro — a $700 camera with interchangeable lens mounts. The appeal? You could slap vintage cinema glass (Leica, Cooke S4) onto a cheap body and get cinema-quality texture at a fraction of the cost of a RED camera system, which can run $7,000+ or $800/day just to rent.

Markiplier saw the potential: the next generation of YouTube filmmakers doesn't have his balance sheet. (Forbes estimated his pre-Iron Lung creator earnings at $30 million per year.) GoPro could democratize cinematic production for smaller creators.

So he bought the stock. Reports suggest he spent $10–12 million to acquire roughly 8% of the company. Then, on August 27th, he posted a glowing sponsored review of the Mission One Pro.

🔄 The Twist: GoPro Gets Acquired

Five days after the sponsored video dropped, Starman Optical announced a merger with GoPro. The deal pays $285 million in cash, retires debt, and leaves GoPro shareholders with 10% of the post-transaction company.

Starman Optical is a holding company controlled by the Tebele family, which also owns consumer accessory brands like Incase, Cygnett, and Griffin. GoPro fits neatly into their portfolio of electronics accessories — similar supply chains, similar distribution.

But here's where it gets messy: Starman also launched a new Photonix division focused on high-speed optical networking equipment for AI data centers.

⚠️ The Backlash: AI Baggage

Suddenly, Markiplier — who is vocally anti-data center and anti-AI — is associated with a company pivoting toward AI infrastructure. He got blindsided. The timeline turned on him. The narrative shifted from "YouTuber supports indie filmmakers" to "Markiplier bankrolls AI compute."

Clarification from Hunter Weiss (friend of the show): Markiplier actually bought shares months ago — the SEC filing just dropped recently, making it look like he bought right before the sponsored post. He also only controls 0.5% of voting power, thanks to GoPro founder Nick Woodman's super-voting Class B shares.

In other words: Markiplier used the product → loved it → bought the stock → promoted it → got blindsided by an acquisition that conflicted with his values.

📈 Why Starman Wanted GoPro

Beyond the brand fit, Starman gets something valuable: a NASDAQ listing. By merging into GoPro and keeping the ticker alive, Starman effectively gains public market access — a bit like a SPAC or reverse merger, but cleaner. The stock is up 33% today.

🏈 The $1.3M Romance Scammer Who Faked an Entire NFL Career

In one of the wildest fraud cases of the year, a 35-year-old man allegedly scammed 26 women out of more than $1.3 million by pretending to be an NFL player for the San Francisco 49ers.

His toolkit:

• Fake Instagram with "Free Agency Year One" and "Super Bowl Highlights"
• Fake bank statements
• Luxury cars
• Dating apps
• An elaborate backstory claiming his family owned Gucci stores on "Saks Fifth Avenue" (a store, not a location)

He even claimed his parents were from Switzerland and were majority shareholders in individual Gucci stores — which is not how that works.

The kicker? Google AI Overviews reportedly picked up his fake claims and started presenting him as an actual 49ers player. He'd show women screenshots of Google saying, "That's Google. That's not me."

He was arrested at an airport with an alleged accomplice, but not before creating an absurd amount of video evidence — including footage of himself walking around shopping in full helmet and gear.

"The question is how quickly Netflix can turn this into an instadoc. There's so much footage. This thing practically edits itself."

Real NFL players are now memeing the story — some have posted videos of themselves casually shopping in helmets as a tribute.

⚡ SpaceX Wants to Make Power Turbine Blades — Big Turbine Is Nervous

Elon Musk confirmed over the weekend that SpaceX will start manufacturing blades and vanes for natural gas power turbines — a move that has rattled investors in the highly concentrated turbine component oligopoly.

Musk tweeted:

"By doing in-house casting at SpaceX, we can accelerate natural gas turbines coming online by up to 18 months, which is a profound game-changer."

📉 Market Reaction: Shares of Howmet Aerospace Fall

Shares of Howmet Aerospace fell more than 7% on Monday before partially rebounding. Howmet and Berkshire Hathaway's Precision Castparts dominate the market for turbine blades and vanes, which are also used in aerospace and defense.

These companies trade at nosebleed valuations:

• Howmet Aerospace: 42x forward earnings
• DPC: 45x forward earnings
• GE Vernova: 33x forward earnings
• Caterpillar: 26x forward earnings

🔬 Why It's Hard (But SpaceX-Shaped Hard)

Making turbine blades is extremely difficult. The parts must withstand extreme temperatures and rapid rotations. The manufacturing process involves:

• Creating a new wax mold from scratch each time
• Growing the blade as a single crystal of nickel superalloy in a vacuum furnace
• Destroying the ceramic cores and wax molds after production to prevent reverse engineering

Even a stray grain or hairline defect means the part must be scrapped. New manufacturing lines can scrap over half their production for extended periods.

Howmet CEO John Plant reassured investors:

"We're not getting distilled. We destroy every piece of evidence that would lead someone to copy us."

But let's be honest: if there's a company built to handle extreme precision, high scrap rates, and rapid iteration, it's SpaceX. They literally grow rocket engines.

Morgan Stanley noted that SpaceX likely won't become a major supplier outside its own use — meaning they're vertically integrating to accelerate AI data center buildouts, not to compete with Howmet directly.

⛵ Bonus: Rotor Sails Are Real and They're Spectacular

Maersk is installing rotor sails on container ships — and no, they don't look like sails. They look like giant smokestacks.

Here's how they work:

• A large cylindrical sail is rotated by an electric motor (not by wind)
• The spinning sail compresses air on one side and thins it on the other
• The pressure difference creates thrust — the same principle behind a curveball in baseball
• In good conditions, rotor sails can produce more thrust than the main engine, saving fuel

It's a fascinating application of the Magnus effect, and it's being deployed at scale by one of the world's largest shipping companies.

🎯 Final Thoughts

From AI safety debates spiraling on the timeline to YouTubers accidentally becoming activist investors, this week's news underscores a few themes:

• AI governance is still a mess. We're adjudicating complex technical debates on X without ground truth.
• Vertical integration is accelerating. SpaceX making turbine blades. Starman merging with GoPro. Everyone's tightening supply chains.
• The internet is undefeated. Whether it's rotor sails, fake NFL players, or Ilya's dachshund timeline, the weirdness never stops.

Stay sharp. The only constant is chaos.