Recent posts by Wired Writers on Bluesky

A year and a day ago, I had surgery to remove a strange mass following a severe bout of sepsis and months of chronic anemia. I didn't know if I would ever run again. Today I ran a 10-mile race and hugged my daughter at the finish line. It was a glorious feeling.

It’s funny how in a 14 day period we here from OpenAI that «this is a critically important moment for cyber defense with AI, there is not much time to act.» and that there is «more than 10% chance AI 'could kill all humans». Yet, we are still waiting for a clarification. www.bbc.com/news/article...

So I told GPT-6 to make me a virtualized simulation of a water treatment plant and now I’m spending a lot of time saying “please document that this is just for defensive purposes” because I realize how bad it looks.

A thought experiment you can do for evaluating whether to use LLMs for advanced mathematics on it’s own: Would you use their proof to develop and test new medical equipment and instruments for measuring partical physics? Or would you wait for the peer-review that is due 2 years from now? #llm #ai

A tiny fly-brain lab on an Adafruit Metro RP2350: real fruit-fly wiring, a separate reward-learning circuit, live graphics, and onboard lights controlled locally with CircuitPython and native C ...

As Fears of A.I. Catastrophe Magnify, Washington Stirs, but Mostly Slumbers www.nytimes.com/2026/09/13/u...

Sign up (or RSS) today to get this week's latest edition: this.weekinsecurity.com 🐈‍⬛

I am sure you have realized that the reason they were able to find the error the LLM had made is because the theorem had already been proven by someone. Yet, OpenAI came out and said that they had an LLM that was better than that research because they had found and error.

#3DPrinted Guardian Robot with LEDs – #Zelda : Breath Of The Wild - Guide: learn.adafruit.com/guardian-rob... youtu.be/ULqynriVAa0 #3DPrinting #Adafruit

Proportionality weighs a restriction against the harm it prevents. If you seriously accept the "AI KILLS ALL HUMANS / WORLD DOOM" risk model, the harm is infinite. Absolutely no restriction is ever disproportionate.

1) Open-weight models are improving with respect to closed frontier models. 2) AI Safety arguments as framed, may help an intervention to keep control.

There is this huge discussion now in Cyber after Sam Altman said: “This is a critically important moment for cyber defense with AI, there is not much time to act.” It’s like the whole AI choir just decided that it’s time to go to church. Nobody is claiming LLM isn’t useful and’t aren’t acting.

Damn that's profound.

Chicago is such a beautiful place.

I just don’t understand how people still call it being on Twitter and sending Tweets? It’s like a large group of the world population is in denial about the last 4 years.

don’t talk to me or my son ever again

Starting my quick talk at @blueteamcon.com unconference in 5 minutes!

Last night I paid $40 (plus tip) for a pizza margherita. I am reporting the pizzeria to the Italian authorities. No pizza should cost that much.

Somebody needs to put "you're paying more for gas so young girls will be forced into genital exams" in a campaign ad.

This is the right take.

It’s 2026 and getting Covid still sucks. If you think you may be sick (with Covid or something else), please please just stay home.

OpenAI’s claim that their AI can be used in mathematical research have been met with heavy backlash. Take f.ex. OpenAI’s claim they disproved the Connes Rigidity Conjecture. J. L. Nielsen from CTP, University of Kansas audited the code line-by-line and discovered that the AI's proof was invalid. #AI

My rough estimate is that around ~9 autonomous AI hacking incidents remain undisclosed/unknown. At least based on the scarce data points we have. techletters.substack.com/p/9-autonomo...