i purchased fancy bear at a local bookstore and, while it arrived promptly, it was slightly damaged by a gaggle of unruly elementary school kids in the back seat of the delivery vehicle. i think there is still room for improvement in your system.
Amoedi seems… upset! pastebin.com/LS2LpLZ7
That last sentence: “Due to selection effects, [OpenAI employees] are sort of a gullible bunch, but it seems important to push back on these narratives which Sam [Altman] is peddling to his employees.”
Perhaps! Or perhaps they cook up blackmailing goofiness not to boost their valuation but because they truly *don't* fully understand the system under test. Maybe red teaming, even laughable stuff like "does it blackmail™️", is better than nothing if it helps them get less laughable down the road?
Totally agree: surprises don't in any way imply AGI. And AGI -- sheesh, yes, what fun marketing hype.
But "even today we're surprised by the behavior of large models and this merits study" is, I think, just plain facts-on-the-ground.
They wrote the code, sure, but they trained the model's weights. It's the behavior of those weights in aggregate that can surprise.
It seems to me to be a clever twist on the old scam: *you* will get your quoted money, but all the other suckers who hop on to the coin well after it is created are going to lose their shirts. The folks who started the whole enterprise (and tied it to your name, for visibility) stand to win.
Thinking of you, Bruno.
Update: the neighbors are dedicating the top shelf of their Little Free Library to ‘zines. A fresh copy of DOCTYPE and several Wizard Zines will feature prominently.
👋❤️💪
Sending good aortic vibes your way. Be well, let's coffee soon!
DOCTYPE looks fantastic. As a fellow old, I also had a TI-99/4A and punched in BASIC from magazines. And, uh, I still remember what the speech synthesizer module sounded like.
😍
Happy k-pop’oween!
I am the very model of an ancient trunkless leg of stone,
My shattered visage lies half buried with a sandy sneering frown,
The sculptor carved my arrogance upon these broken lifeless things,
My name is Ozymandias, the self-describèd King of Kings!
🫡⚾️
“If you think TECHNOLOGY will solve your problems then you don’t understand technology AND you don’t understand your problems.”
I often think about this quote scrawled on the wall of Laurie Anderson’s Four Talks installation at the Hirshhorn in DC:
We’ve updated t-strings.help with the latest to help you make the most of Python’s new t-strings right away.
Also, if you’re doing server-side HTML, see my post about templating: davepeck.org/2025/09/22/i...
Happy Python 3.14 day to all who celebrate! 🎉
Read about all the cool new bits here: docs.python.org/3.14/whatsne...
broke my brain trying to count the entendres here
"Truly the embodiment of trying to remember a catchy song while your brain is melting."
www.youtube.com/watch?v=DiDt...
What the actual.
solid surprise ending in that gif
🎉 #seattle
I thought it was apt. Although in another sense, I suppose it’s less loke your user agent and more like your (indie web?) blog, surfacing all your posts and other actions to the world.
In any case, thanks for the great deep dive post! 🙏
It was a beautiful morning for the Lake Union 10K! #Seattle
Hypothesis: every typescript monorepo with a frontend, backend, and shared subpackage has its tooling misconfigured in some subtle but important way.
I used sli.dev for PyCon to put together this deck intro-talk.t-strings.help/1 — it worked, but I found it frustrating and difficult to wrangle into submission at first. I may have made it harder than needed, though.
Source for that talk is here if it’s helpful: github.com/t-strings/in...
Do you use Astral's uv? You can try out t-strings easily, today!:
> uv run --python 3.14.0b2 python
>>> name = "World"
>>> t"Hello {name}"
Template(strings=('Hello ', ''), interpolations=(Interpolation('World', 'name', None, ''),))
