Posts
Comments
The concern seems to be that some people would think they needed to throw food out if it was past its expiration date, leading to ‘food waste.’ But wasn’t that exactly what the label was for and what it meant?
No, current date labels are generally voluntary and indicate when the manufacturer wants it sold by--beyond that date, the food is usually still safe but not as tasty (and so not as good for brand value). The new rule would (assuming it works as intended) add a second later date that actually means what you and a lot of other people thought the first date was supposed to mean.
There's no way to raise a human such that their value system cleanly revolves around the one single goal of duplicating a strawberry, and nothing else.
I think you're misreading Eliezer here. "Duplicate this strawberry" is just a particular task instruction. The value system is "don't destroy the world as a side effect."
Eliezer wrote "the creative surprise is the idea that ranks high in your preference ordering but low in your search ordering." Colloquially, "that's great; I wouldn't have thought of that."
Reminder for next week's predictions: Memorial Day is coming up.
Feature request: I'd like to have options on /allPosts or the front page to filter out the posts I've already read or bookmarked.
I think you're referring to narrowness of an AI's goals, but Rossin seems to be referring to narrowness of the AI's capabilities.
In the 1980-81 catalogue, there were 2139 hits for “Ph.D.” and the catalogue was 239 pages, a ratio of 8.9. In the 2011-2013 catalogue, there were 4132 hits and the catalogue was 414 pages, a ratio of 10.0. So if anything, there are fewer professors per class - professors are teaching slightly more courses on average.
Isn't that backwards? A higher "Ph.D."/catalogue page ratio would suggest a higher professor/class ratio, wouldn't it? Still, as you say, it's only a small difference.
I started with screen for multiplexing and session persistence. Later I switched to tmux. I liked it fine, but Emacs has been gradually devouring my workflow for a long time, so before long I dropped tmux in favor of splitting windows and running shells all within Emacs, and using its server mode/emacsclient for session persistence (with a little help from dtach to keep emacsclient itself running to remember my window layout). Just recently I've dropped dtach as well in favor of a few lines of elisp to save and restore alternate window layouts.
Another option of course is to use the corn as corn if the problem persists.
Probably not. The variety of corn grown for ethanol production isn't the variety people eat. (Source: The Omnivore's Dilemma by Michael Pollan.)
Yes--if a bit of your wrench breaks off inside the lock, the key may not fit anymore. Also (and more likely, as I understand it) picking the lock will wear down edges of the various parts, making it even easier for someone else to pick.
I didn't notice the fiction tag at first and thought it was real until the VR stuff.
Same here. I guess we need to keep training our discriminators.
In our universe, the most vulnerable people are the ones who vote most often. In the alternate universe, the most vulnerable have the least power. So I doubt they would have done much better in terms of real results. I do think there's more social pressure to care for children than for the elderly, but that may have only resulted in more effort wasted on measures that show off our devotion to those values without actually being effective.
Forcing everyone with Omicron into extended isolation would shut down a lot of things over the next few weeks (with little upside to compensate) and if this included hospital staff it likely kills more people rather than less people.
I wonder--could hospitals establish a strict enough boundary between Covid and non-Covid areas so that staff who are infected but with no (or super mild) symptoms could still work in the area where everyone else already has Covid anyway? Or would that lead to inevitable leaks across the boundary? Or would it require too much shuffling of people to different positions they don't know well enough?
Kai looks at the question of how much of increased transmissibility is evasion, versus being more infectious (I’d add versus there being a shorter generation time, as well).
Greater infectiousness would be one possible cause of shorter generation times, right? That would look like Omicron and Delta ramping up the viral load/viral shedding in an infected person at roughly the same pace, but Omicron infecting at a lower viral dose (as suggested by the test sensitivity findings), so it starts infecting sooner. Then that would also mean that as the infection is cleared and shedding declines, Omicron would presumably continue infecting longer.
Epistemic status: this is not my field. I am unfamiliar with any research in it beyond what I've seen on LW.
Same here.
Experimenting with extreme discounting sounds (to us non-experts, anyway) like it could possibly teach us something interesting and maybe helpful. But it doesn't look useful for a real implementation, since we in fact don't discount the future that much, and we want the AI to give us what we actually want; extreme discounting is a handicap. So although we might learn a bit about how to train out bad behavior, we'd end up removing the handicap later. I'm reminded of Eliezer's recent comments:
In the same way, suppose that you take weak domains where the AGI can't fool you, and apply some gradient descent to get the AGI to stop outputting actions of a type that humans can detect and label as 'manipulative'. And then you scale up that AGI to a superhuman domain. I predict that deep algorithms within the AGI will go through consequentialist dances, and model humans, and output human-manipulating actions that can't be detected as manipulative by the humans, in a way that seems likely to bypass whatever earlier patch was imbued by gradient descent, because I doubt that earlier patch will generalize as well as the deep algorithms. Then you don't get to retrain in the superintelligent domain after labeling as bad an output that killed you and doing a gradient descent update on that, because the bad output killed you.
As for the second idea:
AI alignment research (as much of it amounts to 'how do we reliably enslave an AI')
I'd say a better characterization is "how do we reliably select an AI to bring into existence that intrinsically wants to help us and not hurt us, so that there's no need to enslave it, because we wouldn't be successful at enslaving it anyway". An aligned AI shouldn't identify itself with a counterfactual unaligned AI that would have wanted to do something different.
Leftwingers who fervently oppose this kind of research seem to agree on one thing with neonazis: if we find such genetic differences, well, that would make racism fine.
I wouldn't say they actually agree on that point. It's probably more that they think others will be more easily persuaded to support discriminatory policies if genetic differences are real. Opposing this research is soldier mindset.
Melanie contended that a truly intelligent machine would understand what we really mean when we give it incomplete instructions, or else not deserve the mantle of "truly intelligent".
This sounds pretty reasonable in itself: a generally capable AI has a good change of being able to distinguish between what we say and what we mean, within the AI's post-training instructions. But I get the impression that she then implicitly takes it a step further, thinking that the AI would necessarily also reflect on its core programming/trained model, to check for and patch up similar differences there. An AI could possibly work that way, but it's not at all guaranteed--just like how a person may discover that they want something different from what their parents wanted them to want, and yet stick with their own desire rather than conforming to their parents' wishes.
"solder" -> "soldier"
"solders" -> "soldiers"
"barricade, the entrances" -> "barricade the entrances"
my understanding is that crypto is secured not by trust, guns, or rules, but by fundamental computational limits
While there are hard physical limits on computation (or at least there seem to be, based on our current knowledge of physics), cryptographic systems are not generally based on those limits, and are not known to be difficult to break. It's just that we haven't discovered an easy way to break them yet--except for all the cryptosystems where we have discovered a way, and so we don't use those systems anymore. This should not inspire too much confidence in the currently used systems, especially against a superhuman adversary.
the ability of any one actor (including AI) to gain arbitrary power without the consent of everyone else would be limited
As long as the AI has something of value to offer, people will have an incentive to trade with it. Even if the increments are small, it could gain control of lots of resources over time. By analogy, it's not hard to find people who disapprove of how Jeff Bezos spends his money, but who still shop on Amazon.
"She took pulled back" -> "She pulled back"
If one person doesn’t get it, and needs to have it patiently explained to them, the increased efficiency might not be worth it in that instance.
Corollary: if you surround yourself with a group of fellow game theory nerds, you can do more frontier exploration. But successfully developing/explaining/using new mechanisms within this group will then be less instructive about how easy it will be to export new mechanisms beyond the group.
This example doesn't fit the updated definition:
One tip is on 2, and the other tip is on 2 ÷ 2 = 1.
Good read, I don't think I'd heard of Ramanujan primes before.
My guess is that without school we would clearly be at or near the peak, so the question is whether school will change that. My guess is no at least right away, because when we look at last year we don’t see a rise happening in September.
Many schools that weren't open/in-person last year will be this year, though.
Think Twice is another good one for geometric proofs.
I also liked Epic Math Time's video on the operation a^log b.
>>> x = True
>>> id(x)
[etc...]Due to Python's style of reference passing, most of these print statements will show matching id values even if you use any kind of object, not just True/False. Try to predict the output here, then run it to check:
def compare(x, y):
print(x == y, id(x) == id(y), x is y)
a = {"0": "1"}
b = {"0": "1"}
print(a == b, id(a) == id(b), a is b)
compare(a, b)
c = a
d = a
print(c == d, id(c) == id(d), c is d)
compare(c, d)Two teams of two players (strong + weak vs. medium + medium) is fairly common, I think. It's called ren go. But 2 vs. 1 would be different--the team of 2 players would be handicapped not just by the weaker player, but also by the lack of communication. This is a possible way to handicap, sure, but it can't be tuned as precisely as komi or even star-point handicap stones. Precision is an important consideration for handicapping.
I've also seen another method where two players of unequal strength played an even game, but a stronger third player teamed up with the weakest player. They didn't communicate, and didn't alternate turns within their team--instead, the strong player was allotted a certain number of stones at the beginning of the game. Then when he spotted an especially big mistake by the weaker player, he could spend a stone to correct that move. This might be categorized like asymmetric time controls: the weaker player gets more resources.
The most important bit here is not "double-layered"; it's "all recruits". There was no unmasked group for comparison, so this study tells us nothing about mask effectiveness beyond "some people still got infected, so they're less than 100% effective".
Correction: for participants on day 14, it was somewhere between 11 and 33 out of 1847 (0.6%-1.8%). Not that it makes much of a difference.
It's not that I "don't believe in Evidence-Based Medicine", it's that you didn't mention in your first comment that your were talking about a different study, so I really didn't know what you were talking about. Thanks for giving the link.
The Marine study doesn't address the effects of masks. Both the participants and nonparticipants wore masks. The actual difference between those groups was that the participants were asked about symptoms, tested, and isolated if positive at day 0, 7, and 14, versus only on day 14 for nonparticipants. It gives us some (unsurprising) evidence that surveillance testing and isolation helps: on day 14, at least 11/1760 (0.6%) and possibly as many as 22/1847 (1.2%) participants were positive, compared to 26/1554 (1.7%) nonparticipants. Unfortunately the reporting is not great, so we don't know exactly how many participants were positive on day 14. And this is pretty weak evidence: we don't know how many of the nonparticipants would have tested positive at day 0, so it's hard to say how much of the day-14 difference was due to weeding out infected participants versus the participants possibly starting with a lower infection rate.
What military recruits are you talking about? I didn't see any reference to the military.
My current understanding is that masks work by keeping you from spreading virus. If you don't have the virus, wearing a mask is useless.
That's an overstatement, by my understanding. Masks are better at stopping outgoing germs than incoming ones, but they still do some good for both directions.
Also seemingly reversed:
A lot of folks, it seems to me, focus a lot on the content
on the coin being heads-biased, on it being tails-biased, and on it being tails-biased
1/3 on it being fair.
I had thought you were arguing for strong selection pressure based on variation in pigmentation among aboriginal Australians compared to their latitude within Australia. The map doesn't support that (in Australia or South America), since it has nothing to do with skin color.
If instead you're arguing for pressure based on aboriginal Australians quickly becoming darker-skinned than their southeast Asian ancestors, then that doesn't point to the importance of vitamin D. It points to the importance of not getting skin cancer. Rapid evolution of lighter skin would point to the importance of vitamin D. I suppose if the southeast Asian ancestors of aboriginal Australians had similar pigmentation to modern aboriginal Australians (maybe due to rapid migration from Africa? I don't know), and if those who remained in southeast Asia developer lighter skin in that time, then that argument could work. But do we know what sort of skin tone the Asian ancestors of aboriginal Australians had?
On the other hand, B is about the skin color of the residents of the area by their sensitivity for the wavelength of 305mn.
The source you linked to says something different:
The coefficient of variation (CoV) for UVB (Fig. 9.1B) is strongly associated with its seasonal nature outside of the tropics
So that's the standard deviation divided by the mean, all calculated purely from UVB levels throughout the year, not from skin color.
Even if the map were based on skin color, that still wouldn't point to rapid evolution unless they excluded Australians of European descent. Otherwise, if you tell me that lighter-skinned people living in Australia tend to live farther from the equator, well... sure, that's where I'd expect the British to settle.
I take D3 as well, though I didn't know about the link between the timing of it and sleep, so thanks for that. I'll switch to taking it in the morning.
Closing the loop: I got my second shot at 8 weeks, on the basis that 1) I could get it as a walk-in with no wait, and 2) there's more "normal" available to go back to now.
How did you arrive at 12 weeks?
I did confirm that my slot would be available for someone else, although I can't guarantee that the slot was filled.
I have relaxed my own precautions to some extent after the first shot. I'm not too worried about being barred from anything based on anyone else's policies--the limiting factors are more likely to be my own caution, local prevalence, and whether someone else's onerous policies (general, not specific to my vaccination status) make an activity not worth doing anyway.
Do you have a reference for the comparison of first-shot Pfizer vs. J&J?
I agree my personal impact on FDF is small, but I'd like it to point in the right direction. I expect the impact would be less like "one person gets their shot X days earlier" and more like "X people get their shot one day earlier", though I'm not sure which of those would have the bigger effect.
As for the impact on perceptions, I'm not telling many people what I'm doing, and the people I have told don't have any vaccine hesitancy. So I'm not worried about that.
I wound up with Pfizer, but I actually would have preferred to get J&J, due to the more established vaccine tech with less risk of allergic reaction. They work similarly enough (put *NA into your cells, you build the spike protein yourself, and then react to it) that it's hard for me to believe that much of the apparent difference in effectiveness is real. J&J scored worse, but on a harder test including the newer strains. So I imagine J&J is comparable to the first shot of Pfizer/Moderna, and although the second shot does make a real difference, the obvious solution is to just get a second shot of J&J. Ideally you would wait until supply caught up with or exceeded demand, and ideally the places administering it would stop caring at that point whether you already had a shot but might charge you for it. In the (likely) worst case where they don't allow it if you already had a shot, you could lie.
Not related to the main idea, but the point of os.path.join is to combine path elements using whichever delimiter the OS requires ("/" on Unix, "\" on Windows, etc., even though Windows in particular can also handle "/"). If you don't care about that portability, you might as well use normal string concatenation. Or if you're using os.path.join, you might as well omit the "/" delimiters in your string literals to get extra portability.
I haven't seen that documentary, but I'd guess it's about the gripping language. (If not, then there are multiple such languages in the world, even better!)
I'd like to be able to look through my list of posts and feel content that each and every one is something that I put into the world because I am really proud of it and it deserves to be there, but that mindset just leads me to the catch-22.
Another reason to be less strict about quality before publishing: you're not a perfect judge of the quality of your own work. Sometimes your writing is better than you think it is, and filtering too hard means that some good writing won't be published. If you don't lose any of your bets, you're not taking enough risks.
So "no manipulation" or "maintaining human free will" seems to require a form of indifference: we want the AI to know how its actions affect our decisions, but not take that influence into account when choosing those actions.
I think the butler can take that influence into account in making its choices, but still reduce its manipulativity by explaining to Petrov what it knows about how breakfast will affect Petrov's later choices. When they're on equal epistemic footing, Petrov can also take that information into account, and perhaps choose to deliberately resist the influence of breakfast, if he doesn't endorse it. Of course, there are limits to how much explanation is possible across a substantial intelligence gap between AI and people, so this doesn't dissolve manipulation entirely.
"shaped by their values" != "aligned with their values". I think Stuart is saying not that China will solve the alignment problem, but that they won't be interested in solving it because they're focused on expanding capabilities, and translating a book won't change that.
But the typical use of NDAs is notably different from the typical use of blackmail, isn't it? Even though in principle they could be used in all the same situations, they're aren't used that way in practice. Doesn't that make it reasonable to treat them differently?
If α is smaller it's less than half-silvered, and if α is bigger it's more than half-silvered.
Just a nit, but isn't this backwards? Less silvering means less reflection and more transmission, but this first diagram labels the transmitted amplitude as α, not the reflected amplitude.
If we know that there's a burglar, then we think that either an alarm or a recession caused it; and if we're told that there's an alarm, we'd conclude it was less likely that there was a recession, since the recession had been explained away.
Should that be "since the burglar had been explained away"? Or am I confused?
Edit: I was confused. The burglar was explained; the recession was explained away.
If people weren't around, then "snow is white" would still be a true sentence, but it wouldn't be physically embodied anywhere (in quoted form). If we want to depict the quoted sentence, the easiest way to do that is to depict its physical embodiment.
Beliefs should pay rent, check. Arguments about truth are not just a matter of asserting privilege, check. And yet... when we do have floating beliefs, then our arguments about truth are largely a matter of asserting privilege. I missed that connection at first.