The Future of Humanity Institute closed in 2024 and nothing has replaced it. FHI asked the questions about AI, existential risk and the long-run trajectory of civilisation years before anyone else did, and its corpus is still where much of today's AI safety thinking comes from. I want to build its successor in San Francisco, close to where decisions about advanced AI are being made.
Since September 2025 I have run an FHI paper reading group in San Francisco with excellent co-hosts, held at weekends so that working professionals can attend.
These reading groups are heavily oversubscribed, and invite-only to ensure excellent discussion quality. I plan to train further reading group hosts and scale to meet demand.
We use defamiliarization to let people talk about AI without talking about AI directly: by talking about technology in general, and in more abstract terms. In an NDA climate, this is what lets people reason things through out loud.
I want our venue to be a place where key decision-makers can think hard about the most important cruxes and get unstuck, surrounded by researchers who are good to think with. I expect visitors and participants in our seminars to think critically about what they are doing and why. I take the risk of humanity failing to achieve its full potential as seriously as risks from artificial intelligence development.
We've filed Form 1023 for 501(c)(3) classification, and the IRS has responded requesting more information to consider the determination letter request. We'd like to work with a nonprofit lawyer over the next 2 weeks to reply to this letter (deadline: Dec 2, 2026) & secure 501(c)(3) status, which involves paying them.
Other likely uses of initial funding include:
Cloud/setup/web domain costs
Deposit for our venue
Projector, to facilitate presentations of research
Hiring first employee to complete long-term fundraising
We also have an initial project underway on technological advancements that might facilitate stable equilibria for AI pacing. It focuses initially on the conflict between EAs preoccupied with personal longevity vs. AI pacing enthusiasts. Might we be able to reach 'longevity escape velocity' at lower AI capabilities levels? Is it appropriate to set up isoquants and model longevity as a function of 'AI capabilities' and 'Auxiliary investment'? We'd love to support more small research projects of this flavor, on global priorities research and the order/sequencing in which technologies should be developed. Funding can support early researchers and allow us to make grants to researchers.
I hosted the 90/30 ML paper reading group at Mox in San Francisco to get AI safety researchers more ML-literate, then handed it off successfully. The 90/30 Club has since run 75 sessions over approximately 18 months, with additional branches in NYC, Boston, and at CMU. I worked with Austin Chen on "GiveWell for AI Safety" before departing to pursue technical AI safety research.
$2k BlueDot fast grant
Comments
Austin Chen
4 minutes agoI had the pleasure of collaborating with Lydia for a month last year, for an exploratory research project around "Givewell for AI Safety" and also AI-powered grantmaking. Since then, I've continued to be impressed with her initiatives at building community and supporting sharp, intellectual folks in their journeys through the world. (Whenever I run into a smart young generalist AI safety-ish person, they almost invariably know Lydia.)
She's already taken many concrete steps on making Forward Pass happen, including securing a venue in SF, running well-curated events, and getting advice from many great people (and also, me?). I'm excited to support this with $10k through Sunlight.
Samuel Ratnam
about 1 hour agoI think the field could benefit a lot from the thoughtfulness and sanity that Lydia embodies, especially as we start to enter crazier and crazier territory. I look forward to seeing what comes out of this! (also isn't Forethought kinda an FHI successor? I believe Anders Sandberg is also working on establishing something in this direction)