Talk for The Curve 2025.

About me

• I run Manifund, a nonprofit funding platform & DAF
  • Mostly funds AI safety projects
  • Runs experiments (eg impact markets; quadratic funding)
  • ACX Grants
• Other projects: Mox, Manifest, Manifold
• Think a lot about funding, as funder and grantee; as forprofit and nonprofit

Soon: vast torrents of money available for AI safety and xrisk

• Right now, spending on AI:safety is around 100:1 to 1000:1
  • order of $100Bs/y on AI, $100Ms/y on safety
  • Expect higher ratio towards safety as people wake up
    • Current nonprofit spend on climate change is ~$10B/y
  • Even if ratios stay constant, AI spend is growing
    • eg frontier training run costs growing 2-5x/y, per Epoch
• $Bs of Anthropic & OpenAI equity will be available soon for safety
  • https://www.lesswrong.com/posts/yjiaNbjDWrPAFaNZs/reasons-to-sell-frontier-lab-equity-to-donate-now-rathe

We lack serious financial infrastructure to scale up giving

• Safety spend is ~$100-500m/y, mostly via grants
  • E.g. OpenPhil (over 2-3y): $60m MATS, $25m Bluedot, $25m to FAR
• Pitfalls of grants, as currently practiced:
  • Grants are budgeted to cost, not value
  • Restrictive
  • Very slow
  • “Funder chicken”
• Grant funding today is a command economy