I’d wager that after a certified point funds to lobby on an issue like this have substantially diminished near zero marginal returns. But if you are way behind the first few millions spent may have very high payoffs.

Interesting but that is a rather thin reed to rely on here

OK I think the revised language makes it clerer (see updated version of site ... referring to 'timing gate' etc)

having a think about this.

Update. I've put in some more work into this and I think it's getting better. Please have a look and let me know your thoughts. If it seems valuable, we could go for some more systematic RLHF etc.

We've now updated the page so that you can give quick feedback on the potential impact and 'evaluation relevance' of the suggested research. If you have some familiarity with the research in this area, this could be very helpful for us in tuning and improving this resource.

 

NB -- this is almost entirely AI generated, with some back and forth prompts and corrections

I'm sharing a steelman against a live assumption in Bay/EA/AIS circles: that large AI-lab-adjacent philanthropy is likely to arrive soon enough, and in a sufficiently usable form, that organizations should plan around it.

https://uj-ai-wealth-philanthropy-steelman.netlify.app/

The stronger skeptical case is that IPOs, valuations, pledges, DAFs, and foundation stakes are several gates away from fast, flexible, AIS/EA-directed grants.

The interactive model lets readers vary assumptions about Anthropic valuation, founder ownership, pledge follow-through, employee giving, OpenAI Foundation allocation, lockups, deployment rates, and grantmaker capacity.

Original motivating thread/comment: https://forum.effectivealtruism.org/posts/dtF6wBjH7yBD4kqLz/noah-birnbaum-s-quick-takes?commentId=sGRyGF5wjaaoMFmfK

@Noah Birnbaum 

As a first pass, I asked GPT Pro to consider and model this, and Codex to host it, with interactive BOTEC tools etc.

https://uj-ai-wealth-philanthropy-steelman.netlify.app/

I'm just looking through it now (I'll respond/adapt to hypothes.is comments). Let me know if this sort of thing is useful or annoying.

One statistical/methodological point I’d add (something I always harp on). I don’t think “not statistically significant” should directly cited as evidence for a lack of a difference. If the question is whether mortality differences are small enough to be decision-irrelevant, we’d want something closer to an equivalence test or Bayesian posterior over the mortality difference, plus a welfare model translating mortality causes, morbidity, behavioral deprivation, fear/stress, and transition dynamics into aggregate welfare burden.

A forest plot or explicit meta-analytic summary could also make the cage-free evidence easier to interpret than the table of pairwise significance checks.

Related: I wouldn't always treat small sample sizes or mixed statistical significance as automatically implying “no useful inference.” Small-N studies can be informative if underlying measurement noise is low. For example if I ask 4 people to taste a drink and they all wince deeply in pain and disgust, I'm going to be highly confident it tastes bad. If all 4 smile and praise it, I'll be fairly confident that it's at least tolerable.

McElreath's globe-tossing example illustrates how much we can sometimes learn from small samples.

 (Still, in the shrimp case, it does seem like there is some substantial underlying variation unrelated to the different slaughter methods.)