M

MarcusAbramovitch

2220 karmaJoined

Comments
135

This is generally less than one FTE for an AI safety organization. Remember, there are other costs than just salary.

MATS is spending far more than £500k/year. I don't know how accurate it is but it looks like they might have spent ~$4.65MM. I'm happy to be corrected but I think my figure it more accurate.

The other two things I want to point out are:

  1. It's very tempting to be biased towards "the thing I should be doing is making money". I've seen a shocking amount of E2Gers that don't seem to do much giving, particularly in AI safety. There should be a small anti-correction bias against the thing you should be doing is making money and investing it to earn more money. That looks a lot like selfish non-impact.
  2. £250k/year after taxes and expenses, just isn't that much to donate. I think in the UK (where £250k/year would be paid) would incur income tax of ~35-40% depending on deductions. Let's call it £95k. After say £45k/year in personal expenses (more if you have a family), we are talking about about £110k/year. Invested or not, this just isn't that much money to move the needle on AI safety by enough to write home about. AI governance organizations would very happily have a very good mid to senior operations management roles at EA and adjacent organisations with longtermist focus or other role. These orgs spend  £110k/year like its nothing.

I think this effect is completely overshadowed by the fact if what you are saying is true, we have 5-10 years on the technical alignment/governance of AI to get things to go well.

Now is the time to donate and work on AI safety stuff. Not to get rich and donate to it later in the hopes that things worked out.

From an outsider perspective, this looks like the sort of thing that almost anyone could get started on and I like the phrasing you used to signal that. AI progress moves so fast that you are most likely going to the only one looking at something and so you can do very basic things like 

"How deterministic are these models? If you take the first K lines of the CoT and regenerate it, do you get the same output?"

It's pretty easy to imagine taking 1 line of CoT and regenerating and then 2 lines...

I think a lot of people can just do this and getting to do it under Neel Nanda is likely to lead to a high quality paper.

In common English parlance, we don't preface everything with "I have estimates that state...". 

I don't think any reasonable person thinks that they mean that if they got an extra $1, they'd somehow pay someone for 10 minutes of time to lobby some tiny backyard farm of about 1770 pigs to take on certain oractices. You get to these unit economics with a lot more nuance.

It's eag weekend. I would give at least a week before rushing to a judgement.

First, I want to commend Vetted Causes for doing this. The EA community loves criticism as a concept but rarely likes it when it comes to the object level. This is quite the object level critique and we see how our soldier mindset quickly arose. I'm hesitant to give any pushback on this since I want the cost to doing this work (which is quite hard) to be as low as possible. With that said, I want to 

When it comes to Sinergia, I have quite the strong prior that they would be effective. Sinergia is what you would get if you took THL-style corporate campaigns, put them in the Global South where most of the animals on Earth live, in worse conditions than in the Global North, get way cheaper talent since salaries are lower and put in a kick-ass CEO in Carolina. This is kind of a recipe for an effective charity.

For Vetted Causes, I think your critiques would do better, be more productive for the community and frankly, be better received by the EA community if you would first get feedback from the organization that you are evaluating. You don't even need to change anything based on what they reply. I would recommend you use your judgement in evaluating what they have to say if it corrects any factual errors but the rights to the final wording of the post remain your own. This has three main advantages

  1. It corrects factual errors that might not be public
  2. It gives the organization a chance to give their perspective on things to the public
  3. The community wants this as a norm (for reasons 1&2 and you should want a strong reason to violate a norm) and so your critique will be better accepted if you do this. Again, you don't need to make any changes if the organizations response isn't convincing from your POV.

I again want to commend VettedCauses for doing these. They aren't easy. I'll pre-commit at least $1000 USD to VettedCauses for their next critique (provided it's a serious one) and they run it by the organization in question for feedback.

I just want to add some colour as someone from a Western country (Canada) that has lived a couple years in Latin America. The rule of law is simply followed far less and it is a far lower trust society. However much American companies follow corporate commitments, I expect less of Brazilian ones.

Also, companies in Western countries don't follow the law all the time. In Latin America, the law is a bit of a joke and there is more corruption. The fact that it was already illegal updates me ~0 on what a factory farm in Brazil is doing.

Agree with others. This is fantastic. Learned a lot.

It's a shame and a sign of significant bias in EA that this got so many disagree votes. What are people disagreeing with? Presumably that taxes kill people but the author makes a very simple argument that nearly everyone in EA approximately agrees with. He then goes on to point out cases where the sign could be reversed or approximately 0.

This is lazy voting.

Load more