Lazily pursuing earn-to-give. Very excited about AI Safety, GHD, and the weirder parts of animal welfare.
dan.pandori
Karma: 747
[I did not downvote]
The paper IMO spends too little time on the important question, which is how much impact you get per dollar. The paper instead assumes that you could plausibly hasten developing full aging reversal by 1 second for 5,500$.
For neglectedness, it would be good to discuss current funding levels. Gemini suggested ballpark 11 billion annually.
For tractability, it would be good to discuss particular research lines that have had success. For example, how much have we lengthened mice lifespans? What is our rate of progress there?
Also, many folks on the EA Forum dislike posting links without copying the text to the forum. I mostly agree, although IMO it is fine to do in a ‘Quick Take’.
That’s a fair distinction. And it is also true that this isn’t the main point of your post, so not an issue if it isn’t 100% accurate.
In response to “EA doesn’t recommend becoming a politician”:
80000 does recommend policy work. https://80000hours.org/skills/political-bureaucratic/
EA forum posts making similar recommendations seem common.
It is hard to say if EA-as-a-monolith recommends any particular thing, but I think many EAs would encourage (competent and electable) folks to consider a policy career.
This implies planning to have children is immoral. You get morally penalized for the suffering of your future children, but are not morally rewarded for their joy.
This seems like a large bullet to bite.
I think this is basically the same theory as: https://forum.effectivealtruism.org/posts/je5TiYESSv53tWHC9/utilitarians-should-accept-that-some-suffering-cannot-be-1
Amen that this is both an exciting opportunity, and feels depressing. I didn’t get into EA to buy warm fuzzies, but I do miss them.
EA 1.0 vs EA 2.0 is worse terminology than GHD vs longtermism. GHD and longtermism is more descriptive, and therefore less likely to confuse folks.
SMBC by Zach Weinersmith is doing a great job of conveying AI Safety memes more widely.
Relevant comics: https://www.smbc-comics.com/comic/speech https://www.smbc-comics.com/comic/safe https://www.smbc-comics.com/comic/ai-17 https://www.smbc-comics.com/comic/ai-15
I would love to see his take on an illustrated AI Safety book, like ‘Open Borders’ meets ‘If anyone builds it, everyone dies’.
In very broad strokes, 20% GiveWell & 80% AI Safety. About half of the AI Safety funds are to 501c4s.
Prior years were about 95% GiveWell or GiveWell top charities.
Feel free to DM me and I can share a CSV of my donations (sharing it publicly still feels kinda weird).
Over 40% of my lifetime donations happened this month. I expect to never be able to make that statement again. These are weird times.
For context on why it is extra weird, I’ve been donating since 2015 and have always donated 10% of my salary each year (or more).
I think you’re selling yourself short at 300-500 USD. Gemini estimates 1600-4200 USD (for 3 reviewers total), Opus 400-1000 USD (for a single reviewer spending only 4-6 hours). I endorse those estimates.
Prompt for those curious: If academic peer reviews were compensated at market rate (ie, relative to industry pay for someone with the relevant expertise), how much would it cost to have a typical academic paper reviewed?
I would have found this much more persuasive if you’d tried these services yourself and found them valuable. Without that, my median expectation is that they will do a worse job than Claude Opus 4.7.
Upvoted for recognizing a problem & responding thoughtfully.
This level of aggression towards well-intentioned funders & NGO founders is a net negative. If this kind of discourse were normalized, I think it would reduce engagement with effective charity.
In response to “I hope [the big funders are not] fucking sleeping at night”:
I hope the big funders are sleeping well, getting rest, and engaged with their hobbies. Perpetual terror is not a good mindset for making high stakes decisions.
IMO add it, especially if it bothers you for a given post. Cases are often egregious even when Pangram misses it. I personally feel like these posts end up long winded & eloquent (but empty of surprising insights). I am sad to read what looks to be an effort-post, only to realize it is little more than a prompt.
Alternatively, we should get an emoji react that is just ‘LLM?’
I strongly agree.
Note that even the ‘highly paid’ COO position at CEA is arguably substantially below market rate. A COO in industry would likely get paid double or more.
S risks are a thing. There exist fates worse than death.
‘significantly reduce’ could mean a lot of things. I’m answering as if this reduces absolute X-risk by 20% or more over the next 10 centuries.
The article is good, but the title’s claim is too strong.
Merely knowing that Malawi is a landlocked sub-Saharran African country has huge explanatory power. The question of ‘We don’t know why Malawi is poorer than Rwanda’ seems like a better question (which the article explores).
The pushback against AI led productivity growth also seems comparatively weak. AI is not referenced until the last paragraph, and I don’t think you really engage with what AI makes possible.
This is just annoying because the article is really good, but now I want to argue about the title XD
Yes, if you assume your errors are log normal distributed, you expect to see big errors.
Your simulation says that the range of actual threats is tightly bounded between 10e-5 and 10e-7 (IMO much too small a range). In contrast, your error estimates span 8 orders of magnitude (IMO likely too large a range).
I really think your choice of parameters fully explains your results.