simonw

6 ms·

61.
by simonw 21d ago
In that case the social stigma won't matter. Do you care if the support agents at faceless companies are annoyed with you? A stigma should affect you when dealing with friends, business contacts, or anyone for whom the quality of your
62.
by simonw 21d ago
In that case the startups should be able to defend themselves against the social stigma - "sure, there's a stigma against doing this, but here's why we think it's worthwhile anyway in this case".
63.
by simonw 21d ago
We need a hard social stigma against allowing your AI agent to contact another human being on your behalf.
64.
by simonw 22d ago
> The problem is that using a dramatic rhetorical flourish every third sentence for completely un-dramatic things is fucking tiring . 100% this. It's exhausting reading this stuff, and those dramatic flourishes are a big part of wh
65.
by simonw 22d ago
My understanding is that Irregular were the company that hosted sandboxes to run some of these evals in, and those sandboxes ended up misconfigured. I got the impression that in some cases it was the customer (Anthropic etc) misconfiguring
66.
by simonw 22d ago
Slightly odd update from OpenAI - I think this is the only place they've acknowledged the RubyGems incident: https://openai.com/hugging-face-incident-and-misalignment/ > September 11, 2026: We are investigating
67.
by simonw 22d ago
It's about more than just training data. Every model since GPT-2 had SVGs in the training data, because they all used a scrape of the Web and the Web is full of SVGs. What's unique about Gemini is that they actively worked to get
68.
by simonw 22d ago
Because it's one of the few ways of comparing models that lets you instantly evaluate them visually. That makes it more comprehensible than a numeric score on a benchmark.
69.
by simonw 22d ago
Google Gemini are the only team who have openly had staff deliberately spend time on SVG performance: https://twitter.com/sunjiao123sun_/status/202455551655137292... See also this Jeff Dean tweet showing off their
70.
by simonw 22d ago
I love these.
71.
by simonw 23d ago
One thing that I didn't understand from this article was how you use this deployment platform. Can you tell Claude code for web to build and deploy a web app using it? If so, does that end up with a public URL somewhere? Does that de
72.
by simonw 23d ago
"No systemd. No sshd. No cron. No logging daemon." You can delete that sentence. Why would someone assume sshd or cron in an environment like this? Omit needless words.
73.
by simonw 23d ago
Because you want to use other existing libraries that are written in Rust.
74.
by simonw 23d ago
Yes, Pyodide is almost entirely used in web browsers. I want to be able to run existing code that uses libraries like Pyodide-core in a browser.
75.
by simonw 23d ago
I was a bit nervous about this trend when it started picking up because I care about Pyodide (Python running via WebAssembly) and libraries that use PyO3 might not work in Pyodide. Thankfully that's now been mostly solved - you can com
76.
by simonw 23d ago
TIL Homebrew has its own sandbox mechanism - looks like it's built around their own sandbox-exec wrapper, at least on macOS: https://github.com/Homebrew/brew/blob/d79ef822ab8136e393ed5f...
77.
by simonw 24d ago
> The business could clearly be made profitable – Bending Spoons proved that in a few months. The problem was that Evernote had raised over $200M at a unicorn valuation and investors needed a billion-dollar exit. No amount of expense-cut
78.
by simonw 24d ago
Here's a copy I pasted into a Gist: https://gist.github.com/simonw/3c5a53c012448456729bcfb19b661...
79.
by simonw 24d ago
Yeah, I'm with you on that one. It seems to be a recurring pattern though. The X.ai coding agent harness is 840,000+ lines of Rust - I dug into that here: https://simonwillison.net/2026/Jul/15/grok-build&
80.
by simonw 24d ago
I'm using both extensively at the moment and I've not yet found that either is demonstrably better at specific tasks than the other. They're both beasts! (I've not had Fable do as much work in Blender as Astra yet though
81.
by simonw 24d ago
I'm seeing mixed signals on this one. On the one hand you get stories like Shopify and Cloudflare expanding their intern programs because interns can get so much more done now. On the other hand there seems to be a measurable slowdown
82.
by simonw 24d ago
Here's the one I saw most widely cited: https://neuromatch.social/@jonny/116324676116121930 I wasn't very impressed by that pile-on, personally. Has there ever been a rapidly evolving 1,000,000+ line codebase
83.
by simonw 24d ago
Timing is pretty good for OpenAI right now, given GPT-6 Astra is out and appears to match or even beat Claude Fable 5.1. They might not be as interested in slowing down if Anthropic currently had a significantly better model than them!
84.
by simonw 24d ago
I'm really excited about this library. Wrapture is a new project from Graham, who previously developed wrapt (a 20+ year old Python monkey patching library) and built the first version of the New Relic Python agent - the library that i
85.
by simonw 25d ago
Given this AND the wiki attack, the obvious question right now is how many more incidents like this are out there waiting to be discovered?
86.
by simonw 25d ago
> Our understanding from talking to people in the RubyGems community is that OpenAI never informed them that they were responsible for this attack. I really hope that's not the case, because if it is there are two options, both of t
87.
by simonw 25d ago
Authors Spencer Kitts, Thomas Larsen, Sydney Von Arx - those are the three of the same authors as the Wiki report from last week: https://collusion.wiki/
88.
by simonw 25d ago
> A lot of people survive and find the security they need by solving problems for others. Take that away and they have little need for or access to "the power [of] computers." If LLMs and related technology result in a gutting
89.
by simonw 25d ago
Thanks for clarifying, I genuinely didn't understand your point and I do now. I agree that ChatGPT was launched without having an idea as to what it was for. OpenAI themselves debated internally as to whether it was worth releasing and
90.
by simonw 25d ago
You think those billion people are all addicted to having a made-up friend? I think it's being built to solve problems they have. OpenAI shut down GPT-4o because it leaned too hard into getting people addicted to it.

More ›