Post

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/...

Is this a bit like slime moulds solving mazes? It looks considered but actual fact it’s just brute force approaches until it find the next successful step?

no

Its not brute force, its A*, but A* still isn't cognitive and doesn't understand what any of its output means in the slightest, nor does it have goals

Does that matter though? We have automated systems that can solve extremely complicated problems now. Personally I don't care if they "understand" what they are doing - what's interesting to me is the results

"extremely complicated" ehhhhh, A* isn't solving anything complicated, just parallel. Its actually extremely simple, just a lot of simple things. Metasploit is already an automated system that can solve extremely complicated problems.

The results here are not novel, other than the sandbox authors being incompetent.

It must take significant mental effort to look at this and conclude "not novel" huggingface.co/blog/agent-i...

Why are you reposting the same link I replied to? You think I didn't read it? Jesus christ, you're so busy being smug you can't see that you're being naive.

You started it. "it must take significant mental effort" was not a good faith comment.

That's fair - I wouldn't categorize it as in bad faith because it was what I was genuinely thinking but it was a rude thing for me to say