🚨 What happened at Polymarket today?
• Users reported unauthorized withdrawals. • Polymarket halted trading on its international platform. • The company traced the incident to a compromised third-party provider. • Trading resumed, and Polymarket promised to fully reimburse affected users.
Reports put the suspected attacker’s wallet at ~$2.65M in POL — not a confirmed total loss.
The scary part?
A platform can fix its own security and still be exposed through a trusted third party.
A developer found Claude Code was spending ~16% of their usage on something unexpected:
Writing 3–5 word status updates.
Not coding. Not debugging. Just telling you what the agent is doing.
One user's measurement, not a universal benchmark.
But it raises a bigger question:
How much of your AI budget goes to work — and how much goes to managing the AI?
The AI space is getting louder every day.
More tools. More launches. More hype.
I’m more interested in the people quietly building things that actually work.
If that’s you, I want to know what you’re working on.
No pitches. No guru stuff. Just builders. 🤝
Hey, experts on X! 👋
RAM is still expensive. DDR5 prices have climbed several times over in the past year. When will it actually get cheaper? End of 2026, or not until 2027 or 2028?
AI agents have a problem nobody wants to talk about: unit economics.
Fewer than 10% deliver robust ROI.
Inference costs per agent workflow could rise 5×+ by 2028.
The paradox?
AI is getting smarter. Making money from it isn’t getting easier.
The next AI winners may not build the smartest models.
They’ll make intelligence cheaper to use.
Source: Gartner
Trump is banning the terms “AI” and “Artificial Intelligence” in U.S. government documents.
They’re to be replaced with “Super Intelligence.”
The risk: existing laws, regulations, and contracts are written around “AI.” The rename creates legal ambiguity and makes safety rules harder to apply consistently.
Thieves have stolen over $100M in AI data center equipment since June.
Servers, memory chips, fiber, copper, modular power gear — all targeted while in transit.
Methods: GPS spoofing, ramming escort vehicles (including off-duty cops), hacking freight broker accounts, and AI-generated fake driver’s licenses.
One load of server equipment worth $13M disappeared after the escort was deliberately hit. Another $10M server rack vanished the same way.
The AI buildout made this gear worth more than gold bars on the open road.
OpenAI claims its internal AI just disproved the Exponential Time Hypothesis.
ETH has been a foundational assumption in theoretical computer science for decades. Hundreds of papers treat it as basically true.
The model also claims progress on Kakeya conjecture and parts of Yang-Mills.
Proofs are Lean-verified and top researchers (including people who proposed ETH) are reviewing them.
If it holds — AI just rewrote a piece of the map we thought was solid.
If it doesn’t — still the most serious AI-generated math claim we’ve seen.
Hype or the real thing?
Anthropic filed confidentially for its IPO at a $965B valuation.
OpenAI's last reported valuation: $852B.
SpaceX is targeting roughly $1.75T.
The number that matters isn't the valuation. It's the revenue and margins, which investors won't see until the public filing.
Would you buy at $965B, or wait for the numbers?
Ripple is betting that AI agents will pay each other.
Its XRPL AI Starter Kit lets agents pay for APIs and compute in XRP or RLUSD. Mastercard named Ripple a partner for agent payments.
The risk: the payment flow depends on two systems staying in sync, the website and the blockchain. Ripple hasn't shared real adoption numbers yet.
If agents start paying for their own tools, who signs off on the bill?
Do you feel the danger of AI?
Because Antropik testified about the danger of its AI in the IPO documents.
Gemini 4 Argon is out and I spent an afternoon poking at it.
Long answers are cleaner than before. Some simple questions still come back overcomplicated.
Would you switch your daily work to it yet?