I don't really understand this post; I suspect the LLM has made it a little confused.
You say:
> When we mistake an early-stage research or field-building problem for a last-mile delivery problem, we set ourselves up for disappointment.
Cultivated meat was, for years, implicitly treated as if it was close to a last-mile problem: a few years from cost parity, just needing scale and distribution. It wasn’t. Something similar could happen with welfare tech.
Isn't most of the focus already on development?
> Leafleting was perhaps the original example. In the early-to-mid 2010s, it was often treated as one of the most effective things people could do for animals, based largely on evidence that later looked much weaker than initially believed. The story was wonderfully legible: distribute leaflets, convert people to vegetarianism, reduce animal suffering. Simple, measurable, scalable. However, the reality was messier. The foundational studies were methodologically flawed (e.g., self-reporting on diets leading to social-desirability bias) and long-term effects were unreliably extrapolated. When these flaws were found, there was a rapid downward adjustment, and now leafleting has effectively been fully removed from EA animal work.
I don't see why this was named as a "bubble", my impression is that this was basically good. People thought leafletting was good; people tried leafletting; it wasn't very good; people stopped doing it. That seems very functional to me.
My impression is that bubbles are bad because they are ex-post indications of overcorrections (on entry or exit or both) that have a bunch of harmful externalities - but I think one needs to make an actual argument at the object level for why there could be an overcorrection towards welfare tech rn (e.g. maybe you think that too much funding is going into it too quickly, but are you sure that money would actually be counterfactually spent on other animal stuff? Or maybe you think that people are moving away from other interventions to work on this stuff, but the main people I know doing AW tech dev are unsurprisingly technical people and weren't working on other AW interventions before that, so there's not much of a tradeoff there).
I'm not sure how using Lightcone Commons increases grantmaking-related conflicts of interest compared with the normal LTFF process, but I agree that conflicts of interest are very common in grantmaking.
I really dislike how, after 5 minutes of reading the post, I'm struggling to tell whether:
The post is from METR (pretty sure it's not, but it uses the METR tag, brands itself as a METR thing, and METR has a similarly named internal project)
This post is satire
Whether a human was actually substantively involved in writing this post
Being able to filter out AI-generated content from the front page would be great, but I don't love that people like @Thomas Kwa🔹 will be somewhat obligated to respond to random slop about METR.
I could see that being good. I think I'd still prefer the filter-out-under-n karma option because I understand what's actually going on under the hood. I expect I'd prefer it to the current front page.
It doesn't appeal to me because I want higher info density, and this reduces it - I could see others liking it .
It's not obvious to me that you can't just hide low karma posts as an option - some options: 1. All first-time posts are reviewed by the forum team, who can upvote things and put them above the threshold 2. Make the low karma/low user karma post hiding disabled by default 3. Try out an option for a bit and remove it if it's not working well
When I posted, the front page felt "spammy", with lots of generic, low-quality posts, presumably from first-time posters. I now rarely click on anything below 15 karma, but that was >50% of my frontpage.
Nitpick but I think it’s unlikely that biological humans will completely disappear and I’m an AI safety person. The sota literature is ofc Pantheon season 2 where a few humans choose not to be uploaded.
My impression is that Pangram has very low false positive rates and unclear-to-me false negative rates - so I'd suggest using it to rule things in as AI-generated, but not strongly rule them out.
A little related discussion here https://x.com/caleb_parikh/status/2035434186417262863
Oh lol, I haven't written a post in a while :'). I think it would have been much more obvious if I had seen it for the first time in a blank editor, which is presumably how much people see it for the first time.
I don't really understand this post; I suspect the LLM has made it a little confused.
You say:
> When we mistake an early-stage research or field-building problem for a last-mile delivery problem, we set ourselves up for disappointment.
Cultivated meat was, for years, implicitly treated as if it was close to a last-mile problem: a few years from cost parity, just needing scale and distribution. It wasn’t. Something similar could happen with welfare tech.
Isn't most of the focus already on development?
> Leafleting was perhaps the original example. In the early-to-mid 2010s, it was often treated as one of the most effective things people could do for animals, based largely on evidence that later looked much weaker than initially believed. The story was wonderfully legible: distribute leaflets, convert people to vegetarianism, reduce animal suffering. Simple, measurable, scalable. However, the reality was messier. The foundational studies were methodologically flawed (e.g., self-reporting on diets leading to social-desirability bias) and long-term effects were unreliably extrapolated. When these flaws were found, there was a rapid downward adjustment, and now leafleting has effectively been fully removed from EA animal work.
I don't see why this was named as a "bubble", my impression is that this was basically good. People thought leafletting was good; people tried leafletting; it wasn't very good; people stopped doing it. That seems very functional to me.
My impression is that bubbles are bad because they are ex-post indications of overcorrections (on entry or exit or both) that have a bunch of harmful externalities - but I think one needs to make an actual argument at the object level for why there could be an overcorrection towards welfare tech rn (e.g. maybe you think that too much funding is going into it too quickly, but are you sure that money would actually be counterfactually spent on other animal stuff? Or maybe you think that people are moving away from other interventions to work on this stuff, but the main people I know doing AW tech dev are unsurprisingly technical people and weren't working on other AW interventions before that, so there's not much of a tradeoff there).
I'm not sure how using Lightcone Commons increases grantmaking-related conflicts of interest compared with the normal LTFF process, but I agree that conflicts of interest are very common in grantmaking.
I really dislike how, after 5 minutes of reading the post, I'm struggling to tell whether:
Being able to filter out AI-generated content from the front page would be great, but I don't love that people like @Thomas Kwa🔹 will be somewhat obligated to respond to random slop about METR.
I could see that being good. I think I'd still prefer the filter-out-under-n karma option because I understand what's actually going on under the hood. I expect I'd prefer it to the current front page.
It doesn't appeal to me because I want higher info density, and this reduces it - I could see others liking it .
It's not obvious to me that you can't just hide low karma posts as an option - some options:
1. All first-time posts are reviewed by the forum team, who can upvote things and put them above the threshold
2. Make the low karma/low user karma post hiding disabled by default
3. Try out an option for a bit and remove it if it's not working well
When I posted, the front page felt "spammy", with lots of generic, low-quality posts, presumably from first-time posters. I now rarely click on anything below 15 karma, but that was >50% of my frontpage.
Is there a way for me to filter posts below say 15 karma from my frontage? I couldn’t easily find it on mobile.
Nitpick but I think it’s unlikely that biological humans will completely disappear and I’m an AI safety person. The sota literature is ofc Pantheon season 2 where a few humans choose not to be uploaded.
My impression is that Pangram has very low false positive rates and unclear-to-me false negative rates - so I'd suggest using it to rule things in as AI-generated, but not strongly rule them out.
A little related discussion here https://x.com/caleb_parikh/status/2035434186417262863
Oh lol, I haven't written a post in a while :'). I think it would have been much more obvious if I had seen it for the first time in a blank editor, which is presumably how much people see it for the first time.