Firinn ⏸️
3.7K posts
@firinnish
jack of all trades
"If Anyone Builds It, Everyone Dies" is now out. Please buy a copy before this week ends Sep 20, so the book gets on the bestseller lists. And if you have remaining questions after reading, check the ONLINE RESOURCES, now online. IfAnyoneBuilds.It
The UK edition of my forthcoming book has been finalized ✨
🚨 Urgent: Help fix California's AI Safety Bill Before Wednesday 🚨 SB 53 is an urgently needed bill, but, as written, would only mandate third-party audits of AI companies starting 2030. This is too late. Ask our reps to implement audits starting in 2026.
AGI's potential emergence presents 5 hard problems for U.S. national security: 1-wonder weapons 2-systemic shifts in power 3-nonexperts empowered to develop weapons of mass destruction 4-artificial entities with agency 5-instability rand.org/pubs/perspecti…
You can read our report here: pepfarreport.org. Overall, I came away with an appreciation for why, when you ask people in Congress which foreign aid program they're genuinely proud of, so many of them name PEPFAR. It has done an astonishing amount with 0.08% of the federal budget. I read the procedures for how they make sure clinics aren't overreporting patients with quarterly checkins to see if documented clinic attendance matches reported attendance. I read the accounting procedures. PEPFAR is very serious about oversight. But while some programs get so caught up in documenting what they do that overhead swallows the results, PEPFAR also has really, really good results. Because HIV is such a killer in the regions where PEPFAR operates, you can see PEPFAR's effects directly from national-level mortality records. Calculating the program's lives saved with a differences-in-differences approach from national-level mortality records gets results prettyy similar to the results you get from looking at the benefits of ART. So using several different methodologies, we ended up highly confident that PEPFAR has indeed saved many millions of lives in a highly cost effective way, and our median estimate is about the State Department's. But don't trust us - read our work! The report lays out our assumptions, our questions, our skepticisms, and which papers we drew on. It was a joint project by people coming from a wide range of backgrounds and across the political spectrum. It is a real, sincere effort to take an independent look at this question. It is also a work in progress. It's fact checked and we engaged with external reviewers, but that doesn't mean we're done. We'll be engaging with questions, adding to the FAQ, and making any corrections if they are necessary. The decision of whether to reauthorize PEPFAR and how to improve it is a big one and we want to make sure that everyone, Republican or Democrat or distrustful of both parties, has the tools to make up your mind on this one yourself.
This accepted ICLR workshop is a bad idea.
Will AGI be built by 2030? Jack Clark, Sarah Guo, Dan Hendrycks, Rana el Kaliouby, Eugenia Kuyda, Peter Lee, and Josh Woodward think it likely will be. Ajeya Cotra, Marc Raibert, and Tim Wu disagree.
Sam Altman: "I never pray and ask for God to be on my side, I pray and hope to be on God's side and there is something about betting on deep learning that feels like being on the side of the angels"
