Building a Personal AI Factory
I keep several claude code windows open, each on its own git-worktree. o3 and sonnet 4 create plans, sonnet 3.7 or sonnet 4 execute the plan, and o3 checks the results against the original ask. Any issues found are fed back into the plan template and the code is regenerated.
Bruteforcing the phone number of any Google user
It was going to be tricky to find a display name leak after all that, but eventually after looking through random Google products, I found out that I could create a Looker Studio document, transfer ownership of it to the victim, and the victim's display name would leak on the home page, with 0 interaction required from the victim
Get Up Close to Frank Lloyd Wright’s Unrealized Buildings with David Romero’s Digital Models
“Hooked on the Past emerged from the intersection of two personal passions: the history of architecture and the fascinating world of computer-generated imagery,” Romero says.
When I began this study, I was hoping to find a hidden taboo of holding events on the 11th or typographical bias against the shorthand ordinal. Alas, the reason is far is far more mundane: a numeral
1
looks a lot like a capitalI
or a lowercasel
or a lowercasei
in most of the fonts used for printing books. An11
also looks like ann
, apparently. Google's algorithms made mistakes when reading the11th
from a page, interpreting the ordinal as some other word.
Weather Lab Preview: Cyclones provides early access to experimental cyclone track forecasts from Google DeepMind (GDM) and Google Research's (GR) AI weather models.
You can view our predictions on the website but may not copy, modify, distribute, lease or sell any part of the website (except as permitted by the Terms). You may download and use the underlying data.
It used to be that after I took a half full bottle of ketchup out of the refrigerator I had to turn it over and shake it to get the ketchup down and out. Now the ketchup comes in a container designed to rest upside down with the opening and the twist valve that closes it at the bottom. I don’t know whether it took so long because nobody thought of the idea or because it had to wait on improvements in the valve that let you store the bottle upside down without a risk of the contents leaking.
I wanted to begin a collection chronicling the evolution of the system icons over the years. I've been rolling these out on social media over the past week and will continue to add to and update this collection slowly over the summer. Enjoy!
Adding a feature because ChatGPT incorrectly thinks it exists
We ended up deciding: what the heck, we might as well meet the market demand. So we put together a bespoke ASCII tab importer (which was near the bottom of my “Software I expected to write in 2025” list). And we changed the UI copy in our scanning system to tell people about that feature.
LisaGUI is almost entirely vanilla Javascript. Aside from Gulp.js, which I use as a simple build tool to produce a minified JS file, no third party libraries or frameworks are utilized. LisaGUI contains no code from the Lisa Office System's source code (or any code written by Apple), and doesn't utilize any component of any emulator, like LisaEm or IDLE. It contains no AI-generated code.
Documenting what you're willing to support (and not)
At some point, I realized that if I wrote a wiki page and documented the things that we were willing to support, I could wait about six months and then it would be like it had always been there.
is a personal project designed to visualize the movement and trends of Hacker News.
The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin. The ultimate reason for this is Moore's law, or rather its generalization of continued exponentially falling cost per unit of computation. Most AI research has been conducted as if the computation available to the agent were constant (in which case leveraging human knowledge would be one of the only ways to improve performance) but, over a slightly longer time than a typical research project, massively more computation inevitably becomes available.
Top 100 domains: google.com, googleapis.com, cloudflare.com, apple.com, …
How well do you know JavaScript's Date class?
new Date("0")
The string "0" is interpreted as the year 2000, not as a timestamp!
Nabeel Qureshi has an amazing post called Reflections on Palantir, where he ruminates on what made Palantir special. I wanted to do the same for OpenAI while it's fresh in my mind. You won't find any trade secrets here, more just reflections on this current iteration of one of the most fascinating organizations in history at an extremely interesting time.
First up: children’s books. After filtering the data to focus on animals who were explicitly gendered (she/her or he/him) and appeared in at least 10 different books, only a few animals were more consistently gendered female: birds, ducks, and cats.
The rest—frog, wolf, fox, elephant, dog, monkey, bear, rabbit, mouse, and pig—skew male.