Comments
162
Headline finding: I audited 17 AI Safety Talent programmes. Zero of 17 have published any comparison group, rejected-applicant follow-up, matched control or randomisation. Not one. Every programme that mentions a counterfactual does it by asking participants to self-report.
Background
At least $70 million...
You'll probably get more informative comments if you include a bit more context around this ratings sheet. What led you to select these particular charities to focus on? What are some of the ways your model differs from those of GiveWell/ACE for charities they also evaluate?
Also, cells with comments in them have only a tiny tag to indicate this, which can be hard to see on a sheet this expansive and colorful. I'd recommend listing citations on a separate sheet for ease of use (or something along those lines).
Also do you know where I can find the ACE models?
The charities I chose wasn't a very principled decision, I just used the Givewell #1 (Malaria Consortium) and compared it to a couple other charities I thought could be more effective. The model is really just a basic attempt at comparing animal charities to human charities which is surprisingly something I haven't seen anyone try yet. There's also the utility number and duration of suffering I use to try to get a good grasp of how bad the life is for a chicken/fish