Non-EA interests include chess and TikTok (@benthamite). We are probably hiring: https://ââmetr.org/ââhiring
Ben_Westđ¸
Karma: 15,730
Thanks Vasco, I hadnât seen that. Do you know if anyone has addressed Nathanâs âComparative advantage means Iâm guaranteed work but not that that work will provide enough for me to eatâ point? (Apart from Maxwell, who I guess concedes the point?)
why are there fewer horses?
+1 to this being an important question to ask.
+1 to maintaining justification standards across cause areas, thanks for writing this post!
Fwiw I feel notably less clueless about WAW than about AI safety, and would have assumed the same is true of most people who work in AI safety, though I admittedly havenât talked to very many of them about this. (And also havenât thought about it that deeply myself.)
Is the amount which has been donated to the fund visible anywhere?
Sorry, I donât mean models that you consider to be better, but rather metrics/âbehaviors. Like what can V-JEPA-2 (or any model) do that previous models couldnât which you would consider to be a sign of progress?
What are examples of what you would consider to be progress on âeffective video predictionâ?
LLMs have made no progress on any of these problems
Can we bet on this? I propose: we give a video model of your choice from 2023 and one of my choice from 2025 two prompts (one your choice, one my choice) then ask some neutral panel of judges (Iâm happy to just ask random people in a coffee shop) which model produced more realistic videos.
Thanks for sharing this!
Iâm saying that the authors summarized their findings without caveats, and that those caveats would dramatically change how most people interpret the results.
(Note that, despite the âMITâ name being attached, this isnât an academic paper, and doesnât seem to be trying to hold itself to those standards.)
I agree that the authors encourage this misreading of the data by eg saying â95% of organizations are getting zero returnâ and failing to note the caveats listed in my comment. If you believe that this statement is referencing a different data set than the one I was quoting which doesnât have those caveats, Iâd be interested to hear it.
95% of the time, AI fails to generate significant revenue for businesses that adopt it
I think this is a misreading of the study, though the article you link seems to make the same mistake. Hereâs the relevant graph:
The finding is that 5% of all companies (not just those that have adopted AI) had an executive who reported âa marked and sustained productivity and/âor P&L impactâ of a task-specific GenAI.
I think a more accurate summary of the paper is something like â80% of LLM pilots are reported as successful by executives.â[1]
- ^
Assuming that all successful implementations were preceded by a pilot; the paper doesnât seem to say
- ^
Congrats Aaron!
When I worked at CEA, a standard question I would ask people was âwhat keeps you engaged with EA?â A surprisingly common answer was memes/âshitposts.
This content has obvious downsides, but does solve many of the problems in OP (low time commitment, ~anyone can contribute, etc.).
+1, this seems more like a Task Y problem.
My impression is that if OP did want to write specialist blogposts etc. they would be able to do that (probably even better placed than a younger person, given their experience). (And conversely, 18 year olds who donât want to do specialist work or get involved in a social scene donât have that many points of attachment.)
I use DoneThat and like it, thanks for building it!
Thanks for writing this upâI think âyou donât need to worry about reward hacking in powerful AI because solving reward hacking will be necessary for developing powerful AIâ is an important topic. (Although your frame is more âwe will fail to solve reward hacking and therefore fail to develop powerful AI,â IIUC.)
I would find it helpful if you reacted more to the existing literature. E.g. I donât think anyone disagrees with your high-level point that itâs hard to accurately supervise models, particularly as they get more capable, but also we have empirical evidence that weak models can successfully supervise stronger models and the stronger model wonât just naively copy the mistakes of the weak supervisor to maximize its reward. Is your objection to this that you donât think that these techniques wonât scale to more powerful AI, or that even if they do scale it wonât be good enough, or something else?
I interpret OPâs point about asymptotes to mean that he indeed bites this bullet and believes that the âcompensation scheduleâ is massively higher even when the âinstrumentâ only feels slightly worse?
In his examples ( and lexically ordered) there is no âmost intense suffering which can be outweighedâ (or âleast intense suffering which canât be outweighedâ). E.g. in the hyperreals no matter how small or large .
S* is only a tiny bit worse than S
In his examples, between any S which canât be outweighed and S* which can, there are an uncountably infinite number of additional levels of suffering! So I donât think itâs correct to say itâs only a tiny bit worse.
Thanks for writing this Seth! I agree itâs possible that we will not see transformative effects from AI for a long time, if ever, and I think itâs reasonable for people to make plans which only pay off on the assumption that this is true. More specifically: projects which pay off under an assumption of short timelines often have other downsides, such as being more speculative, which means that the expected value of the long timeline plans can end up being higher even after you discount them for only working on long timelines.[1]
That being said, I think your post is underestimating how transformative truly transformative AI would be. As I said in a reply to Lewis Bollard who made a somewhat similar point:
If Iâm assuming that we are in a world where all of the human labor at McDonaldâs has been automated away, I think that is a pretty weird world. As you note, even the existence of something like McDonaldâs (much less a specific corporate entity which feels bound by the agreements of current-day McDonaldâs) is speculative.
But even if we grant its existence: a ~40% egg price increase is currently enough that companies feel cover to be justified in abandoning their cage-free pledges. Surely âthe entire global order has been upended and the new corporate management is robotsâ is an even better excuse?
And even if we somehow hold McDonaldâs to their pledge, I find it hard to believe that a world where McDonaldâs can be run without humans does not quickly lead to a world where something more profitable than battery cage farming can be found. And, as a result, the cage-free pledge is irrelevant because McDonaldâs isnât going to use cages anyway. (Of course, this new farming method may be even more cruel than battery cages, illustrating one of the downsides of trying to lock in a specific policy change before we understand what the future will be like.)
- ^
Although I would encourage people to actually try to estimate this and pressure test the assumption that there isnât actually a way that their work can pay off on a shorter timeline.
- ^
Your answer is the best that I know of, sadly.
A thing you could consider is that there are a bunch of EAGxâs in warm/âsunny places (Ho Chi Minh City, Singapore, etc.). These cities maybe donât meet the definition of âhubâ, but they have enough people for a conference, which possibly will meet your needs.