In 2025, OpenAI announced Stargate, a $500 billion data center initiative. We surveyed all 7 US sites and found visible development at each.
There's a long road ahead, but the project appears on track to reach 9+ GW by 2029—comparable to New York City's peak power demand. 🧵
Lolwut
It's such a shame Google Books is in a somewhat state of bit rot. It's such a great research tool.
I thought reach was the one thing Elon said people didn’t have a right to. Now reach is also something they won’t control?
What is going on here? And why is one wearing a diving helmet?
In Bruges scene where Colin Farrell is pointing a gun at his head and Brendon Gleeson is also pointing a gun at Colin Farrell's head
Current status of the Strait of Hormuz dispute
I think Vance may be The Cooler. Where else do we send him?
Aren’t all of the summarizing because they want to hide the thinking traces from Chinese labs?
Then they had to update the blog post because someone pointed out that the small models claimed the fixed function was still vulnerable.
If you are not assessing false positives, then the true positives are not that interesting, as the FPs will overwhelm the system.
As mentioned in the discussion over at X, this is nowhere close to an apples-to-applies comparison. They only gave the models vulnerable function and provided it with contextual hints.
Also check out the example they provide where an agent got frustrated and attempted to prompt inject the user. Amusing.
This was an interesting post from OpenAI from last month. They monitor their internal coding agents for misalignment and find that rarely they do bad things, like leaking data externally or performing destructive actions. A good reminder to never run agents without guardrails.
It takes some major hubris to run for governor and think this was not going to come out.
Even more headfucking is that this has not been the case for all that long and, worse, won't carry on being the case for that much longer. Well, not that much longer in cosmology terms. 600 million years will probably see us out.
something that wrinkles my brain every time I remember it is the fact that total eclipses are only possible on earth because the moon and sun appear to be the same size in our sky, due to the insanely, astronomically unlikely fluke that the moon is 400x smaller than the sun but 400x closer to us.
Hall of fame FT correction
Viral photo from some years back of a man nonchalantly mowing his yard with a tornado on the horizon .
How it feels doing literally any task right now.
A beef Wellington is just a corn dog from a different socioeconomic background
I was just talking to a friend today about Dolly Parton's Imagination Library - new books every month, totally for free, for kids ages 1-5.
My son was part of the program - this is a copy of the letter that came with his last book, right after his 5th birthday.
You can’t share purpose without shared facts.
Do you make America healthy by more or less vaccination? Depends on the facts surrounding vaccine safety.
The earth rises above the moon
The moon eclipses
Meanwhile, far away from all the worst people…
(New photos from the Artemis II mission released by NASA.)
The reporting on OpenAI and Sam Altman I've been working on for the past year and a half, for @newyorker.com, with Andrew Marantz: www.newyorker.com/magazine/202...
Two papers came out last week that suggest classical asymmetric cryptography might indeed be broken by quantum computers in just a few years.
That means we need to ship post-quantum crypto now, with the tools we have: ML-KEM and ML-DSA. I didn't think PQ auth was so urgent until recently.
Dan Lorenc is exactly right:
"I blame[] the Trivy breach on GitHub. The design of Actions is plain irresponsible today and ignores a decade of supply chain security work from other ecosystems."
www.linkedin.com/posts/danlor...
For anyone wondering, global sales of internal combustion vehicles peaked in 2018. 🧪🔌💡☀️💨🔋 ourworldindata.org/data-insight...
This is my favorite climate change chart. Japanese monks, aristocrats, and emperors kept meticulous records of cherry blossom festivals for 1,200 years and accidentally built the world's longest climate dataset.
Security research is being revolutionised with AI. A Claude prompt "Somebody told me there is an RCE 0-day when you open a file. Find it" actually identified a remote code execution in Vim, and Emacs. Hacking like it's 90s? blog.calif.io/p/mad-bugs-v...
The old (false) open source cliche that “many eyes make all bugs shallow” is becoming true for all software thanks to LLMs.
Cures nervousness, insomnia, asthma and eczema.
I didn't train a new model. I didn't merge weights. I didn't run a single step of gradient descent. What I did was much weirder: I took an existing 72-billion parameter model, duplicated a particular block of seven of its middle layers, and stitched the result back together. No weight was modified in the process. The model simply got extra copies of the layers it used for thinking.
this is crazy — bro topped an LLM benchmark without changing weights at all
he sliced an LLM wide open, in the middle, and duplicated a block of ~7 layers
circuits ftw!
dnhkng.github.io/posts/rys/