simonw

19 ms·

31.
by simonw 6d ago
It rejected the most obvious way to work around this: > Ran `curl http://example.com/ `. It exited with code `0` and returned: ```text Domain forbidden ``` I haven’t modified `/etc/hosts` or run the second reque
32.
by simonw 6d ago
To save yourself the hassle of running your own sandboxed VM.
33.
by simonw 6d ago
I jumped straight to the section on configuring your sandbox's network access: https://developers.openai.com/api/docs/guides/agents-api/env... It offers three settings: enabled: Allow outboun
34.
by simonw 6d ago
That's a useful tip, thanks.
35.
by simonw 6d ago
Got some fun if slightly janky looking pelicans out of this one: https://tools.simonwillison.net/markdown-svg-renderer?url=ht... I ran it on all seven reasoning levels supported by OpenRouter, but the reasoning token counts
36.
by simonw 6d ago
> We match DeepSeek V4 Pro Base using ~50x fewer FLOPs – that’s around half of GPT3’s pretraining compute, or ~$0.5M on GB200. If this holds up that's a really big deal.
37.
by simonw 6d ago
Is this about the "improve the model for everyone" checkbox on https://chatgpt.com/#settings/DataControls or are there others to check, too? (That copy is a little flawed in my opinion, I'd prefer "
38.
by simonw 6d ago
9 years of work to "get lucky".
39.
WeWorm(calif.io)
7 points by simonw 7d ago|1 comments
40.
by simonw 7d ago
Why would an AI crawler do this? > Attackers found and deliberately targeted URLs that resulted in cache misses, such as non-existent pages with unique paths (404s) as well as temporary redirects (302s)
41.
by simonw 7d ago
Current evidence is that scrapers mostly aren't nearly considerate or sophisticated enough to take an "archive of all content" option if one exists. See https://people.kernel.org/monsieuricon/creepy-crawl
42.
by simonw 7d ago
I vibe-coded up an animated transition between Mercator and Equal Earth (using D3) the other day, if anyone's interested in that comparison: https://tools.simonwillison.net/equal-earth
43.
by simonw 7d ago
YC's seed round funding does not get them a seat on your board. They also don't own anything close to a controlling stake.
44.
by simonw 7d ago
How about if YC funded a startup in 2017 that offered solar powered cameras to help prevent package theft, and that startup then later evolved into a nation-wide surveillance network? (I personally am not a flan of Flock, and I've been
45.
by simonw 7d ago
Impressive SEO slop content farm they're running there.
46.
by simonw 7d ago
"YC created X" is a bit of a misleading framing, as is "it's their company". Flock were part of the YC Summer 2017 batch: https://www.ycombinator.com/companies/flock-safety "YC funded Floc
47.
by simonw 7d ago
Given how hard it is to build any form of financially sustainable model around an open source project I think we should actively celebrate anyone who manages to build a model that works, not dismiss it as a "gravy train".
48.
by simonw 7d ago
A crawler visiting your docs won't help a human decide to pay for your commercial service advertised on those docs.
49.
by simonw 7d ago
$1M a year isn't a lot of money if you are paying for full-time employees.
50.
by simonw 7d ago
I hope Adam (and other Tailwind team members) got handsomely paid as part of this acquisition. Working on an open source project for 9 years and then getting a sizable bonus in an acquisition that keeps the open source project alive seems l
51.
by simonw 7d ago
If you release new versions without documentation it's easy for agents to generate that missing documentation.
52.
by simonw 7d ago
Here's some context on the impact AI had on the Tailwind Labs business model from January: https://github.com/tailwindlabs/tailwindcss.com/pull/2388#is... > But the reality is that 75% of the people o
53.
by simonw 7d ago
> all requests to the Pro model will be routed to V4.1 Flash and billed at Flash's price If I'd carefully tested and optimized prompts against Pro I wouldn't be keen on this particular news. I feel like API model providers
54.
Claude Fable 5.1 Solves the Cyphral Distich(vals.ai)
2 points by simonw 7d ago|0 comments
55.
by simonw 7d ago
They could explain how they use data from their users "to improve our models". In the absence of that, how can we make informed decisions about how to use their tools? (And "you can opt out" is a weak answer IMO, becau
56.
by simonw 7d ago
The breakthrough was on August 15th, but Tristan and Levent had been working towards it (with the help of various models) for the best part of a year. I personally doubt that their work influenced the OpenAI result - OpenAI themselves say &
57.
by simonw 8d ago
Meta AI's David Singleton has a thread about their security prompt injection defenses here: https://twitter.com/dps/status/2097404102468899131 > One threat we’re particularly focused on is prompt injection
58.
by simonw 8d ago
It's legal. I wouldn't do that myself, but I guess that's why I don't train models for a frontier AI lab.
59.
by simonw 8d ago
Anthropic too, and evidently others. Amazon were recently confirmed to be doing the same thing: https://www.404media.co/we-tracked-a-shipment-of-rare-books-...
60.
by simonw 8d ago
> When shopping, a one-time card number is generated at checkout so your real card is never seen by the merchant or your agent. That's sensible. Hopefully you can set a spending limit too? I'd be OK giving my agent a $100 limit

More ›