Discussion about this post

User's avatar
Joe Hajj's avatar

Peter, your work is always as good as it gets. Thank you for everything you do!

Hoping for an update from you on METR’s time horizon eval results for Opus 4.5 (4 hrs 49 mins) whenever they publish the full results. If the 7 month doubling time holds, we’d get near week long tasks by September 2028. METR sets their AI R&D automation threat model (which is >10x’ing an AI researcher’s speed) at 40 hrs

Nicholas Halden's avatar

Who is your best bet for the next leader of Venezuela?

25 more comments...

No posts

Ready for more?