Our new paper is out in AI & Society: "The Necessity of AI Audit Standards Boards,"
By myself, Sammy Martin, Mark Bailey, @contact.ms, & Ross Greutzemacher.

We argue that without a board creating independent audit standards, AI auditing is broken by design.
link.springer.com/article/10.1...

They were translating the proof of Fermat's Last Theorem into Lean; an intermediary step wasn't compiling. They noticed a lemma was wrong.

(They have already fixed the proof but this is very funny! In some sense, the theorem has only been proven this year.)

What move will make our world look more like a world that cooperates, coordinates, and survives? Come up with that move and make it.

If you have any influence, any capital, or any capacity to make the governments look up and listen to the scientists, the best time to spend it is yesterday; the second best time is now.

Sadly, the release of o3 makes it very clear there’s little time left. I expect to probably see at least one Christmas after this one; but I don’t know how many more we’ll have.

But until o3, I was uncertain. I expected the progress to go approximately a specific way, but I wouldn’t have been too surprised to be wrong.

We’re now in the endgame.

After GPT-2 came out, I predicted weakly general AI with a median <2030 on Metaculus. After that, everything seemed to go as expected, with only minor surprises or minor absences of progress.

White chocolate, too:

I just made pure chocolate with colors!

If your technology is distinguishable from magic, it is not advanced enough!

No dyes! Just sugar + cacao + tiny amounts of milk and soya + science 😎

A Facebook post saying that I’m bisexual, published on December 8, 2021.

It’s been three years since I came out!

I’m glad I did it while still in Russia, two months before the war made me leave the country: it’s awesome to know I had enough courage for it and to have learned that even in Russia, I was surrounded by people who loved and accepted me for who I am.

Only on lesswrong