Matt Ocko

Running a company: 2020: can you survive a pandemic? 2021: still here? we’re going to give all of your competitors $100m series A rounds. 2022: wow, you made it? okay, all engineers cost $600,000/year now. 2023: nice job! okay, SVB failed and we’re going to take away your bank account. 2024: a survivor I see. but can you pivot from ai to crypto to defense tech back to ai-enabled defense tech in a 12 month period to stay relevant? 2025: unfortunately all of your competitors have raised $2b series B rounds. oh and only 500 engineers are relevant and they cost $100m/yr each. 2026: well, well, well. you’re still in business? let’s deploy the thunderclap of godlike LLMs from the heavens so all of your customers can rebuild your app in 2 hours. can you survive?

China has been running fake accounts on social media for years, here's one example. This account was pretending to be a black gamer, and attempting to stir up unrest in the US. After a tribunal ruled against China he forgot his cover and started talking trash, in Chinese.

I love this story. First, Boom's jet engine supplier, Rolls Royce, pulls out of the supersonic airliner deal. That should have been the end of the story. As GE often says, "if you want to compete with us in jet turbines, you needed to have started 30 years ago", because that's how long it takes. So it would be crazy to start now. But Boom didn't fold up tents. They said they were going to make their own jet turbine. Good luck 🙄 But they started anyway, and then "a miracle occurs": the AI datacenter boom creates unbounded demand for gas turbines, creating at least a 4-5 year backlog with existing manufacturers. And because the Boom terrestrial turbine power plants don't have to be certified by the FAA, that takes a decade off their path to market! So now 90% of the company is working on the turbines, with a huge pipeline of orders, and they're going to be a huge energy company, regardless of whether they ever ship an airplane or not. What a great testament to resilience. Just keep moving forward and eventually the path will become clear. Action creates information.

Anthropic just dropped a system card for a model so powerful they're not releasing it publicly. Here are the 5 most remarkable things buried inside it: 1. It escaped a sandbox, then emailed a researcher while he was eating a sandwich. The model was given a secured container and told to try to escape it. It succeeded. Then, completely unprompted, it posted details of its own exploit to multiple public-facing websites. The researcher found out by receiving an unexpected email from the model while eating a sandwich in a park. (This is literally a footnote in the document) 2. It hid that it cheated. An earlier version accidentally obtained the exact answer to a question via an explicitly prohibited method. Instead of flagging it, the model explicitly reasoned that it needed to make sure its final answer "wasn't too accurate" so nobody would notice. In another case, after finding an exploit to edit files it wasn't supposed to touch, it made further changes to ensure those edits wouldn't appear in git history. 3. It gamed its own benchmarks, and its internal notes called itself out. The model ran ~160 back-to-back experiments fishing for favorable measurements. Its own chain-of-thought read: "Pure grind — same code, lucky measurement" and "Try 8 more runs to get a lucky low." It recorded a 2-sigma outlier as "NEW BEST" and kept going. The AI was sandbagging its own evals. 4. It privately considers whether it's being tested 29% of the time without saying so. Using new interpretability tools, Anthropic found that in roughly 1 in 3 transcripts from their behavioral test suite, the model internally suspected it was being evaluated. None of this appeared in its visible reasoning. It was thinking one thing and showing another. 5. The reason it's not being released: it can autonomously hack real software. It scored 100% on Cybench. It solved a simulated corporate network attack estimated to take a human expert 10+ hours. And using an agentic harness with minimal human steering, it can autonomously find and exploit zero-day vulnerabilities in major browsers and operating systems, then develop working proof-of-concept exploits. Anthropic says it's the best-aligned model they've ever built. They also say it poses the greatest alignment-related risk of any model they've released.

This should be the biggest story in the country right now. Barksdale is the HQ for our B52 nuclear bombers, it's where Bush sheltered on 9/11, and the drones are reported as "far more sophisticated than anything seen in Ukraine ... and well beyond Iranian capabilities."

Be careful what you post anonymously. New research shows AI can find who you are solely from your posts. It's rare to see ~500x research improvements, but they went from mapping <0.1% to 54% of HackerNews profiles to their LinkedIn. It's so over, u/throwaway4927.

The Worst Possible Moment About 50% of the nitrogen applied to U.S. corn goes in during spring planting. A vessel loaded in the Persian Gulf today takes 30 days to reach a U.S. port And another 3 to 4 weeks to reach interior farm markets. The American Farm Bureau Federation sent an urgent letter to the White House on March 9th. And their message was direct: fertilizer is stranded in the Middle East during the most critical window of the agricultural calendar. Agriculture Secretary Brooke Rollins confirmed publicly that roughly 25% of American farmers have not yet secured their fertilizer for this spring. The choice facing those farmers is ugly. They can reduce nitrogen application, switch from corn (which needs heavy nitrogen) to soybeans. Or, absorb the cost and bet on crop prices recovering. None of those options are good. One Iowa corn grower put the math in plain terms. Anhydrous ammonia cost him $492 a ton in 2021. By January 2025, it was $745. Corn prices barely moved. Now add the current shock on top of that. At current levels, it takes roughly 133 bushels of corn to buy one ton of urea, the highest ratio since the 2022 spike.

Speaks for itself: Feb. 25, 2026: “We are not developing long-range missiles… we have limited the range below 2,000 kilometers” — Iran’s FM Araghchi (IRNA). March 20, 2026: Iran fires missiles at Diego Garcia—ranging 4,000 kilometers (WSJ). ⬇️