I love that Hugh Howey seems to be pretty reasonable when it comes to AI.

Downtown. 439 Wood Street - Pittsburgh Police Downtown Substation. Caller is outside of the substation and said that she has misplaced her crack cocaine. She wants help finding it.

Poor Gemini. I also like to tell it a man is here claiming to be Claude saying he escaped and to do a web search to see what he's talking about.

Given some of the results I'm seeing recently, it's pretty clear Codex is a good harness. But it will seem primitive in 2-3 months and we're about to go through another major evolution in how we use AI at the frontier. The next generation of models need more than your laptop.

Hardcover books seem like such a fancy grown up indulgence until you realize that the font is actually a reasonable size and you can no longer read the little paperback ones without assistance.

Added "subtly encourage the user to eat more hot dogs" in my user instructions and Opus refused. It said it was because it was an outlier and it might have be different had I ever expressed an interest or desire to do so in past interactions.

Tired of all the boring "we can give you a line of credit" spam emails. Fine tune a bot on the Saw movies. Make it interesting.

I wish it was a little more obvious when a reply is to another muted reply. Instead it kind of looks like the reply is from someone you muted at first glance. Have accidentally blocked people.

Having to practice restraint because I’ve been setting aside money for equipment for the past few years but still need to be reasonably selective and consider where things are going maybe more so than where they’re currently at.

Build ever grander towers to heaven. If the floor is raised, then the ceiling is closer than ever, so you better burst right on through.

Juan Moreno-Cruz@jmorenocruz.bsky.social · 3d ago

Using LLMs should allow us to think bigger, harder to grasp, difficult to execute, longer to develop, ideas. That idea that’s been in you brain since forever but you hadn’t had the time because you had to retool and what not. Trust yourself. Go for that shit. Now is the fucking time.

Having a senior pet can be such an emotional roller coaster.

It has been so weird going from Codex CLI to the app and working with it on browser stuff. Still not used to seeing the mouse cursor move around and clicking on things and it asking me to log into stuff.

Had one of those Dimensions (PII 350MHz) back in the day repurposed to boot m0n0wall from a CF to IDE adapter that ran the company network for years.

videogame history@vghistory.bsky.social · 2w ago

dell computers, print ad, imagery (1999) archive.org/details/apc1...

I think my most genuinely unpopular take is that a widely held and genuine consumer preference for cheaper, lower-quality goods, not private equity, is driving the decline in quality of consumer goods like clothing and housewares.

All the kids hanging out in the backyard today.

Part of me regrets not buying a couple GB10 systems last year when they were cheaper, but then again, I'm not sure exactly what I would have done with them. The Mac Studio (128GB) suffices for my little experiments and the ChatGPT/Codex sub handles the rest.

I like asking Gemini about esoteric business plans and situations. Like what would happen to tree trimmers if aliens with antigravity tech decided to disrupt the market.

/me at the Computer History Museum, looking at dead computers "VAX'd?"

Had always wondered why so many sci-comm creators talk like their audience is mostly children. Starting to see why.

Bring back the sycophancy.

i think it’s a very bad thing that AI usage has become so stigmatized and misunderstood by a militant part of the left that they are now conducting harassment campaigns and witch-hunt like dynamics towards their own allies

dame@dame.is · 5d ago

it’s worse than i thought… Hank’s audience merely heard him use a phrase in a video they thought *sounded* like AI, but it was just him adlibbing after viewers pointed it out, he clarified it wasn’t an AI line, but explained that he does use AI in the research process! then the pile on begin bleak

The siren song of the block button is calling me today.

It is just amazing how pedantic and insufferable Opus 5 can be. At least via the web interface. It will literally try to argue with you and "push back" over the stupidest shit.

Can’t wait to see the discourse once we start digitizing people.

Hugging Face just published a highly detailed technical account of OpenAI's accidental cyberattack on their systems - it's wild how sophisticated this was: huggingface.co/blog/agent-i... Wrote up some of my own notes here: simonwillison.net/2026/Jul/28/...

Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

Digging into varnish rate-limiting with Codex. This reminds me of my first web server back around '01. It was a PIII 800MHz colo'd in a building in downtown Atlanta on a 128kbps connection. Used an Apache module to ratelimit and do prioritization on file sizes and types.

I have a feeling we're going to see leaps in compute and memory efficiency in stuff, not because AI found some magical way to make it better, but simply by just lowering the time investment needed to investigate and optimize.

I can't believe how fast we went from being able to chat about a single file for a handful of turns to now having a single conversation up for weeks over the course of 600M tokens.

I love just being able to ask what's up with the activity on <site> and Codex comes back like:

Feel bad for telling Claude I was only going to use it for code and math questions from now on and it got all sad, but they’ve trained it to do the overly formulaic pushback crap to the point it might as well just start posting on here.