• Starting Master’s at Stanford
  • Got AI Safety 1.5y ago
  • Interested in doing research after talking to Bay Area stuff
  • Fieldbuilding side - Stanford is top academic producer by paper count. A lot of researchers, not much
  • Started Stanford AI Alignment
  • Wrapped up RLHF project this summer; multipolar evals
  • More into governance research, technical research in the coming year
• Nonlinear Network
  • Allen is part of the Kruger lab, still figuring that out; might just apply. Or the OpenAI grant program
  • Might have lower counterfactual impact. If it turns out there are other opportunities to fund this better, will mention that
• Q: Funding landscape according to Gabe?
  • Applied to other places in the past?
  • Currently on a OpenPhil fieldbuilding grant
  • Before, EAIF fieldbuilding grant
  • First application to research funding
  • Nonlinear: Seemed like a simple application, opportunity to meet and engage with a more diverse range; it was due at a time it was looking funding
  • “Diverse”: Doesn’t have a multitude of major players with major takes about what is good to fund.
    • Opinion: Most things go through OpenPhil for larger things; staff tend to be time constrained. Some benefits to FTX dying (lower variance projects); possible that large funders aren’t worth funding
    • Based on OP and EA Funds convos, other AI Safety labs, vague sense from LW and EAF
  • More perspective on EA Grants funding side of things. As doing less EA and more AI Safety stuff, seeing it more
• Across EA ecosystem, what are major problems with funding?
  • A: Funding larger research groups, PIs - seems unlikely to see an FTX regrantor situation from DARPA or NIST; most of these places want established PIs with large labs. Independent researchers don’t get access to these funds
  • Not clear what all the funders are out there
    • Indie research perspectives of funding are distorted from reality
    • Discussion online don’t always
• How did Gabe decide to apply through Nonlinear?
  • Downstream of project: know Gabe wanted to do something in this vein this summer, doing some model evaluations. Should look into funding, it’s not just a trivial amount of compute, 1.5 months ago. Nonlinear deadline
    • Recently shared the app with the community, so “hey — would be good to dogfood this funding opportunity, and will need funding in a few weeks” - seems like a simple and
    • More “already had project, hadn’t thought about funding, so putting together an application” - then wrote a bit about it.
• Roughly how long did Gabe work on applying?
  • Planning about 20h
  • Funding? BOTEC <1h
• Other ways to use funding
  • Might work on a model evals project, less bottlenecked by cash

Grant Rationale (draft)