Non-EA interests include chess and TikTok (@benthamite). We are probably hiring: https://ââmetr.org/ââhiring
Ben_Westđ¸
Karma: 15,843
My read of their phase one plan is that they were intending to get these pretty low quality tabloid stories as a springboard to getting higher quality stuff. Maybe that was a bad plan, but the fact that the bad tabloid articles were in fact bad tabloid articles doesnât seem to discredit that?
Is this just a statement that there is more low-hanging fruit in safety research? I.e., you can in some sense learn an equal amount from a two-minute rollout for both capabilities and safety, but capabilities researchers have already learned most of what was possible and safety researchers havenât exhausted everything yet.
Or is this a stronger claim that safety work is inherently a more short-time horizon thing?
This is great.
Thinking through my own area of Data Generation and why I would put it earlier than you, I wonder if there should be an additional factor like âImportance of quantity over quality.â
For example, my sense is that having a bunch of mediocre ideas about how to improve pretraining is probably not very useful to capabilities researchers. But having a bunch of trajectories where an attacker inserts a back door is pretty useful to control research, even if these trajectories are all things that would only fool amateurs and not experts. And if you are doing some sample inefficient technique, you might benefit from having absolutely huge amounts of this relatively mediocre dataâa thing even relatively mediocre models could help with.
Thanks Sarah! Is something written up about the CEA donation system? Iâm surprised that thatâs a priority, but obviously know zero details.
Note that some of the founders have pledged to donate 80% of their equity.
Yeah, I agree that the hypothetical EA seems less like a âradicalâ abolitionist (for some definition of âradicalâ).
âIts mandate was to promote gradual emancipationâ
Your hypothetical EA sounds kind of like an abolitionist to me :)
The Society formed a ways-and-means committee to deal with the difficulty that more than half of the members, including Troup and Jay, owned slaves (mostly a few domestic servants per household). The committee proposed a plan for gradual emancipation: members would free slaves younger than 28 when they reached the age of 35, slaves between 28 and 38 in seven yearsâ time, and slaves over 45 immediately. This proposal failed however, and the committee was dissolved.
https://ââen.wikipedia.org/ââwiki/ââNew_York_Manumission_Society
Thanks for finding this and writing it up! And thanks to FRI for updating their report.
I thought this was an interesting point, thanks for writing it.
Thanks!
I feel confused about this response. Youâre asking for people to give you examples of a thing occurring, Iâm asking by what date range you wish to see examples in.
What time frame are you interested in? E.g. if someone says that they have <30y timelines today, would that meet your criteria?
Thanks! Perhaps I phrased this poorly; a person being a patient or not isnât the relevant factor, itâs whether or not they are licensed. E.g. if you look at the FDA authorization for the first product it says:
The ContaCT mobile application is intended to be used by neurovascular specialists, such as vascular neurologists, neuro-interventional specialists, or users with similar training who have been pre-authorized by their Healthcare Organization or Facility.
Iâm actually not sure whether one could generously interpret âsimilar trainingâ to include e.g. radiology technicians. They wouldnât be allowed to make diagnoses, and my guess is that the government would not look kindly on a rad tech saying something like âIâm not diagnosing you with a stroke, but the AI thinks youâve had one, wink, wink,â but Iâm not sure. Perhaps someone with more legal experience could chime in.
In any case, Iâm skeptical that a business would want to run that malpractice risk (particularly since, as mentioned above, insurance wouldnât reimburse them for doing so).
And yes, I agree that this probably means these products arenât more clearly safe and effective than e.g. eyeglasses (where businesses are analogously legally prohibited from giving glasses to patients without a licensed human optometrist first performing an exam). Itâs just worth considering that this is a very high bar![1]
- ^
Although I think maybe itâs more accurate to just say that medical device authorization is based on a bunch of factors that are largely unrelated to the safety and efficacy of the product. E.g. I think thereâs no one who believes that cigarettes are safer than eyeglasses, despite them being available OTC.
- ^
I doubt that there are surveys of when people stayed home. You could maybe try to look at prediction markets but Iâm not sure what you would compare them to to see if the prediction market was more accurate than some other reference group.
Thanks for collecting this timeline!
The version of the claim I have heard is not that LW was early to suggest that there might be a pandemic but rather that they were unusually willing to do something about it because they take small-probability high-impact events seriously. Eg. I suspect that you would say that Wei Dai was âlateâ because their comment came after the nyt article etc, but nonetheless they made 700% betting that covid would be a big deal.
I think it can be hard to remember just how much controversy there was at the time. E.g. you say of March 13, âBy now, everyone knows itâs a crisisâ but sadly âeveryoneâ did not include the California department of public health, who didnât issue stay at home orders for another week.
[I have a distinct memory of this because I told my girlfriend I couldnât see her anymore since she worked at the department of public health (!!) and was still getting a ton of exposure since the California public health department didnât think covid was that big of a deal.]
Congrats Samantha and the AIM team!
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.
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?)
Hmm. I understood them to be saying they (semi-) voluntarily scaled back before phase 2 was complete, so we canât read that much into the fact that phases 2â3 (where the high quality journalism happens) didnât happen. Maybe I misunderstood?