⛓️

EA needs better software

Main points

  • On the margin, the amount of money EA spends on software is like an order of magnitude too low
    • Examine: SFF, FTX, OP grants
    • Software made up like 10% of the last funding round of SFF
  • Effective Altruism is good at producing ideas but not as good at executing
    • Silicon Valley startups are highly competent
    • “Professionals talk logistics”
  • Research writes onto paper; software writes onto minds.
    • E.g. social networks, video games shape the way people think
    • Think about how much time you spend in front of a screen. The ability to change what gets brought up changes how productive you are
  • Just like research is a public good, software is a public good
    • So we should expect it to be underfunded, especially on EA teams
  • With a good team, software development can be very cheap in terms of time
    • We built the MVP of Manifold in 3 weeks
  • Why does EA software suck? (or does not exist)
    • Very surprising given overlap in EA & engineering mindset
    • Also surprising given that roughly all the money in EA came from one of 3 tech initiatives (FB/Asana, Skype, SBF)

Notes

  • By software I tend to mean “apps” not “sites”
    • Distinction: “apps” are focus on interaction (Gmail); “sites” focus on static content (80k hours). Definitely a spectrum though.
    • Example EA apps: EA Forum; Ought; Metaculus; Manifold; Guesstimate/QURI
    • For comparison: Number of people in university groups, look at funding tables

Meta Proposals

  • YCombinator for EA software
    • Standard terms: $400k for 7% of teams
    • Fly teams in to a physical location; weekly demo days
    • FTX Fellowship, ACX Grant are steps in this direction but not wholly done
    • Incubating new project founders
  • Software consultancy for EA orgs
    • Team of 5-10 software eng who are loaned out to different teams to execute on short-term projects
    • Allows for lots of benefits of software teams
      • Code review
      • Infrastructure build-out
      • Consistent ladder
  • Directing top software engineers into working on EA projects
    • A lot of software engineers are already sympathetic to EA
    • But the “work” done in EA seems to be mostly by researchers
      • e.g. in terms of full time headcount
    • Reach out to EA groups at Google/Facebook etc
    • Match them up with open headcount roles

Project Proposals

  • Utility: Make independent funding much easier to get
    • Lots of funding programs at the moment; hard to navigate
  • Shaping the way people think: Actually effective education
    • Not: “here’s a badge for reading this article”
    • E.g. Anki flashcards; see Andy Mutsachak and Quantum Country
  • Shaping the way people think: microcovid on steroids
    • Micromorts, micromarriages, microjobs
    • Help people quantify risks and rewards of different actions
    • e.g. “A burger costs 10 micromorts, aka 2 hours off your life”
  • Shaping the way people think: Games that teaches EA principles
    • Factorio teaches you how to automate
    • Hanabi taught me to see from other’s perspectives
    • E.g. calculating cost effectiveness for different interventions
    • E.g. managing longtail risks
    • E.g. empathizing with the life of a factory farmed chicken
  • Making it easier to move money around
    • Actually going from “yeah this person should have a lot of money”

Cons

  • Software development has high opportunity costs; that is, founders and employees can make a lot of money in other roles.
    • Google fresh grad makes $200k, at senior (e.g. 3-4 years of experience) makes $400k
      • Software engineers can reasonably earn to give
    • This is evidence that software dev is very valuable, and thus worth EA funding.
      • Unless for some reason it’s more valuable to the for-profit world than nonprofit. I doubt this though.
    • Founder: could aim to be Dustin Moskovitz/SBF instead of Holden Karnofsky??

Specialized software vs general software

  • One possible thesis: “EA doesn’t have to fund software; EA’s software needs are quite minor, and general purpose software like Google Docs and Slack will suffice!”
  • Counterpoints:
    • Specialized software is really, really great; highly leveraged, acts as a force multiplier in different industries. Eg. Dentists.
    • Funding models for typical SV software require a strong growth story; so if something would be especially useful for the population of 3000 EAs, it might not get funded
    • Google Sheets and Google Forms are actually quite unusable, by people. They trade off ease of setup against unfriendliness of site
      • Pick on Nonlinear team for their Airbnb house thing

Importance

Neglectedness

Tractability