Dashiell

I don't think there's a world where mathematicians aren't pissed about this. 1. Dumping everything is what the advisory group said 2. Just yesterday people were mad that Ant paid two experts to digest the 3SUM result What is the middle path where everyone is happy? github.com/…

GitHub - openai/math github.com

I think the recommendations are the only thing that has been published agmai.org/general-sep29/

general-sep29 agmai.org

I would be very interested to see a survey! I know more people who are some combination of pissed and grieving than who are particularly excited, but you prob have a bigger network than me. I am also basing this a bit off the writing here proofsandprompts.com

Proofs and Prompts proofsandprompts.com

This Erdos problem was solved by GPT 5, the technique was new and interesting in its own right, and the prompter was a high school kid with no math training www.erdosproblems.com/forum/thread...

1196 Discussion Thread | Erdős Problems erdosproblems.com

AI safety folks should be way more worried about concentration of power than the proliferation of powerful models

I kinda think that grappling with modern ML turns you at least a little insane. If you aren't at least a little crazy it's because you have not engaged with it enough

It would be wild if AI can find more galactic algorithms. My sense is that up till now computer science has treated them like a pathology, but if they are everywhere we might have tk take them seriously

As recently as 2024 I would say to other ML researchers that I thought LMs were overrated. I was wrong

Now that I have a more normal ML job that I don't need to learn more math for I can spend my time learning algebraic geometry

Do we have any idea whether or not it was regular agentic RL pressure that has caused the most recent crop of alignment issues? For a while it seemed from the outside like the Askell-ite idea of "raise Claude to be good" was working really well