philh

Karma: 7,683

philh Jan 9, 2024, 7:22 PM
2 points
0
on: Dating Roundup #2: If At First You Don’t Succeed

as the (let’s face it bad, but watchable if you don’t care, and worth watching if you need to learn this and other important related lessons) movie Ghosted illustrates

Note, wikipedia lists three movies named Ghosted. I guess you probably mean the 2023 one?

philh Jan 7, 2024, 4:47 PM
4 points
2
on: If Clarity Seems Like Death to Them
But … “I thought X seemed Y to me”^[20] and “X is Y” do not mean the same thing!
And it seems to me that in the type of comment Eliezer’s referring to, “X seemed stupid to me” is more often correct than “X was stupid”.
Argument for this: it’s unlikely that someone would say “X seemed stupid to me” if X actually didn’t seem stupid to them, so it’s almost always true when said; whereas I think it’s quite common to misjudge whether X was actually stupid.
(“X was stupid, they should have just used the grabthar device.” / “Did you miss the part three chapters back [published eight months ago] where that got disabled?”)
So we might expect that “more often true ⇒ less information content”. We could rewrite “X was stupid” to “this story contained the letter E” and that would more often be true, too. But I don’t think that holds, because
- “X seemed stupid” is not almost-always true, unlike “this story contained the letter E”;
- But if someone said “X was stupid” I think it’s almost-always also the case that X seemed stupid to them;
- And in fact people don’t reliably track this distinction.
I think people track it more than zero, to be clear. But if I see someone say “X was stupid”, two prominent hypotheses are:
1. This person reliably tracks the distinction between “X was stupid” and “X seemed stupid”, and in this case they have sufficient confidence to make the stronger claim.
2. This person does not reliably track that distinction.
And even on LessWrong, (2) is sufficiently common that in practice I often just rewrite the was-claim to the seemed-claim in my head.
(Actually, I think I’m imperfect at this. I think as a rule of thumb, the “was” claim updates me further than is warranted in the direction that X was stupid. My guess is that this kind of failure is pretty common. But that’s separate from a claim about information content of people’s words.)
So I think Eliezer is giving good advice for “how to be good at saying true and informative things”, as well as good advice for “how to discuss an author’s work in a way that leaves them motivated to keep writing”.
What links here?
- philh's comment on If Clarity Seems Like Death to Them by Zack_M_Davis (Jan 15, 2024, 1:10 PM; 2 points)

philh Jan 7, 2024, 2:50 PM
2 points
0
on: If Clarity Seems Like Death to Them
I have an inalienable right to talk about my own research interests, and talking about my own research interests obviously doesn’t violate any norm against leaking private information about someone else’s family, or criticizing someone else’s parenting decisions.
I think you’re violating a norm against criticizing someone’s parenting decisions, to the extent that readers know whose decisions they are. I happen to know the answer, and I guess a significant number but far from a majority of readers also know. Which also means the parent or parents in question can’t easily reply without deanonymizing themselves, which is awkward.
This isn’t to take a stance on what you have a right to do or should have done. But I think it’s false to say that you obviously haven’t violated the norms you mentioned.

philh Jan 2, 2024, 10:57 PM
3 points
0
on: E.T. Jaynes Probability Theory: The logic of Science I
Spoiler alert, using a lot of tedious math (that we honestly do not fully follow)
I had trouble with this too. I asked on math.stackexchange about part of the derivation of the product rule. Apparently I understood the answer at the time, but that was long ago.
the most extreme is the ‘A implies B’ which many people consider equivalent to ‘B implies A’ where the correct logical rule is ‘not B implies not A’. Jaynes argues this is an example of Bayesian reasoning. For example if all dogs are mammals (deduction) it also means that some mammals are dogs, concretely if only 20% of animals are mammals this information increased the likelihood of dogs by a factor of 5!. A implies B means that B increases the likelihood of A by a factor of 5.
Had trouble following this because “animals” are mentioned only once, so to elaborate: we’re taking the background info to be X = “this thing is an animal”, along with A = “this thing is a dog” and B = “this thing is a mammal”. Then $A \Rightarrow B$ is $P (B | A X) = 1$ , and if $P (B | X) = 0.2$ then
$P (A | B X) = \frac{P (B | A X) P (A | X)}{P (B | X)} = 5 P (A | X)$
Jaynes’ very physicist-oriented example concerns the emission of particles from a radioactive source and a sensor that can detect some portion of these particles. The radioactive source emits on average $s$ particles per second drawn from a Poisson distribution and the sensor detects particles with an Bernoulli rate of $ϕ$ . So if $ϕ$ is 0.1, the sensor picks up 10% of particles over the long run, but if you have just $n = 10$ particles, there is no guarantee that it will detect exactly 1 particle. Similarly, even though we might have $s = 100$ particles per second, any given second might have more or less than 100 particles emitted. This is where it gets complicated. If you use the MLE estimate, you will always get $ϕ s$ particles as your estimate for each second of counts, because MLE ‘assumes’ that the 10:1 particle relationship is fixed and thus ignores the Poisson source variability. So let’s say you have a counter with $ϕ = 0.1$ and have observed a count of 15 particles on this sensor for some second. How many particles, $n$ , have originated from the source during this second? MLE will get you 150 particles, as described above. But Jayne’s robot gives us 105 particles. What? This is a HUGE difference! This example also surprised experimental physicists. The reason the robot gets 105 and not 150 is because the source has lower variability than the detector, so a high number is weak evidence of an above average number of particles.
Had trouble following this too. I thought we were trying to estimate $s$ and/or $φ$ . But that’s not it; we know s = 100 and φ = 0.1, and we know we detected 15 particles, and the question is how many were emitted.
And if we do an MLE, we’d say “well, the number-emitted that gives us the highest probability of detecting 15 is 150”, so that’s the estimate. We’re throwing away what we know about the distribution of how-many-emitted.
And I guess we could instead ask “what fraction of emitted particles did we detect?” Presumably then we throw away what we know about the distribution of what-fraction-detected, and we say “well, the fraction-detected that gives us the highest probability of detecting 15 is 0.15″, so that would be the MLE.
Which gives us another way to see that MLE is silly, because “what fraction did we detect” and “how many were emitted” are the same question, given that we detected 15.

philh Dec 31, 2023, 9:12 AM
3 points
on: Expected utility and repeated choices
There’s an unstated assumption here that you start with $0. Suppose instead you start with $0.5: then $E (U (a)) = \sqrt{1} = 1$ while $E (U (b)) = \frac{2}{3} \sqrt{1.5} + \frac{1}{3} \sqrt{0.5} \approx 1.05$ . So if you play game (a) first, you’d then prefer to play game (b) second.
But this doesn’t fully resolve the question, because you’d still prefer (b, b) over (a, b).

philh Dec 30, 2023, 4:17 PM
11 points
0
in reply to: Valentine’s comment on: Here’s the exit.
It’s not possible to take the downvotes as a signal of this if downvotes get used for a wide range of things.

Perhaps not in general, but I think it’s often pretty clear. Like you’ve already said “I’m guessing the intention was to punish me for not caring”, and yes, I think you’re right. Seems to me the signal was recieved as intended.

Although if the person disagrees with whether it was bad, and the answer to that disagreement is to try to silence them… then that seems to me like a pretty anti-epistemic norm. At least locally.

Well, if someone comes here arguing for flat-earthism, I’m probably going to downvote without bothering to read their arguments. Is that anti-epistemic? Maybe, I guess? Certainly yes, if it turns out that the earth is flat (and that their arguments are correct). And “this practice isn’t anti-epistemic as long as we only dismiss false ideas” is, um. Nevertheless, I endorse that practice.

If someone comes around here calling people names, and we downvote that rather than checking in “hey are you doing this because you think name calling is good actually? Would you like to dialogue about that?” is that anti-epistemic? Again, maybe yes? But I endorse it anyway.
The dynamic would be pretty simple:
- After I downvote, I skim the replies to see if someone else already explained what had me do the downvote. If so, I upvote that explanation and agree-vote it too.
- If there’s no such explanation, I write one.
Easy peasy.
I do not consider writing these explanations to be easy.

I seriously doubt the number of things needing downvotes on this site is so utterly overwhelming that this approach is untenable.

I can think of a few places we might disagree here: how many things deserve downvotes, how costly it is to explain them, how realistic it is for people to pay those costs. I’m not super enthusiastic about trying to drill down into this, though.

But I also think I’m less optimistic than you about the benefits of doing it. I can think of multiple conversations I’ve had where I wanted people to change what they’re doing, I explained why I thought they were doing something bad, and they just keep on doing it. You yourself seem to understand what it is that many people dislike in many of your posts and comments, and yet you keep doing the thing. Surely there are cases where it does help, but I think they’re a minority. (It seems plausible to me that the helpful cases actually do get explained more often than others. E.g. if someone explicitly asks why they’re getting downvoted, that’s evidence they’re interested in improving, and also it makes them more likely to get an explanation.)

Another thing worth mentioning is that reacts reduce the cost of explaining downvotes. I dunno how much they’re used, since I mostly use GreaterWrong which doesn’t (yet?) support them. I believe they were only added to this post later, so they wouldn’t have been helpful at the time. But yeah, if a comment gets downvoted a bunch with not even any reacts explaining why, that seems not ideal.

philh Dec 27, 2023, 12:36 AM
4 points
0
in reply to: Valentine’s comment on: Here’s the exit.
But it’s clear that some of what I said was heavily downvoted because I took a stance people didn’t like. Saying things like “Yep, I could have phrased this post in a more epistemically accurate way… but for this post in particular I really don’t care.”
Well, that particular comment had a lot of other stuff going on, and yes I think it’s a kind of comment that doesn’t belong here and no I don’t particularly feel like explaining that.
But also, yeah, I do kinda feel like “downvoting people when they admit they did something bad” is a thing we sometimes do here and that’s not great incentives. If someone wants to avoid that kind of downvote, “stop admitting to the bad thing” seems like an obvious strategy. Oops! And like, I remember times when I asked someone a question and they got downvoted for their answer, and I did think it was a bad answer that in a vacuum deserved downvotes, but I still upvoted as thanks for answering.
I’m not sure it’s so bad though. Some things that mitigate it as a strategy:
- “This person strategically fails to answer certain questions” is a thing it’s possible for someone to notice and point out.
- Someone might not have realized the thing they did was bad-according-to-LW, and the downvotes help signal that. (Maybe better to instead upvote the admission and downvote the thing they did? But that’s not always a thing that can be downvoted, or downvotes might not be specifically targetable to make it clear “this thing you did was bad”.)
- If someone did a bad thing and doesn’t care, maybe we just don’t want them here. Downvotes probably marginally push them away, as well as marginally push them towards not-admitting-things. Notably, I feel like we’re more likely to downvote “I did a bad thing and don’t care” than “I did a bad thing, oops, sorry”.
- Sometimes someone might take “not being able to say a thing” as a cost, and prefer the downvotes over the silence.
In general it seems like a hard problem, and it’s not clear to me that downvoting this kind of thing is a mistake.
I’d also really like to see a return of the old LW cultural thing of, if you downvote then you explain why. There are some downvotes on my comments that I’m left scratching my head about and going “Okay, whatever.” It’s hard for downvotes to improve culture if the feedback amounts to “Bad.”
I think there’s currently too many things that deserve downvotes for that to be realistic.

philh Dec 23, 2023, 6:01 PM
3 points
1
on: Bayesian Injustice
Notably, this effect could mean you differentially hire presties even if they’re slightly worse on average than the normies. I feel like practicing my math, so let’s look at the coin example.
If you flip once and get H, that’s an odds ratio of $\frac{2 / 3}{1 / 3} = 2$ for a heads-bias, which combine with initial odds of $1$ to give odds for [a $2 / 3$ heads-bias] of $2$ which is a probability of $2 / 3$ , and your expectation of the overall bias is $\frac{2}{3} \cdot \frac{2}{3} + \frac{1}{3} \cdot \frac{1}{3} = \frac{5}{9} \approx 0.556$ .
Now suppose you flip twice and get HH. Odds for heads-bias is $4$ , probability $4 / 5$ , which gives expectation $\frac{4}{5} \cdot \frac{2}{3} + \frac{1}{5} \cdot \frac{1}{3} = 3 / 5 = 0.6$ .
But suppose you flip twice and get HH, but instead of the coins being a mix of $2 / 3$ and $1 / 3$ heads-biased, they’re $p$ and $1 - p$ , for some known $p$ . The odds ratio is now $o := {(\frac{p}{1 - p})}^{2}$ , probability of [a $p$ heads-bias] is $\frac{o}{1 + o}$ and the expected bias is $\frac{o p}{1 + o} + \frac{1 - p}{1 + o}$ . I tried to expand this by hand but made a mistake somewhere, but plugging it into wolfram alpha we get that this equals $5 / 9$ when
$p = \frac{17 \pm \sqrt{17}}{34} \approx {0.62, 0.38}$
Where the two solutions are because an equal mix of $p$ and $1 - p$ biases is the same as an equal mix of $1 - p$ and $p$ biases.
So you’d prefer an HH with a most-likely bias of $0.63$ over an H with a most-likely bias of $0.66$ .

philh Dec 21, 2023, 12:39 PM
11 points
−1
in reply to: habryka’s comment on: Nonlinear’s Evidence: Debunking False and Misleading Claims
Alice claimed to be making significant income from her side business.

To clarify further, my read of things is that you think the inaccurate claim would be
- Alice was in fact making significant income from her side business.
but that you wouldn’t dispute
- Alice claimed to NL that she was making significant income from her side business.
Is that right? Or do you additionally think the second is inaccurate?

philh Dec 15, 2023, 8:50 PM
16 points
10
in reply to: lsusr’s comment on: How do you feel about LessWrong these days? [Open feedback thread]
I think I’m less open to weird ideas on LW than I used to be, and more likely to go “seems wrong, okay, next”. Probably this is partly a me thing, and I’m not sure it’s bad—as I gain knowledge, wisdom and experience, surely we’d expect me to become better at discerning whether a thing is worth paying attention to? (Which doesn’t mean I am better, but like. Just because I’m dismissing more ideas, doesn’t mean I’m incorrectly dismissing more ideas.)

But my guess is it’s also partly a LW thing. It seems to me that compared to 2013, there are more weird ideas on LW and they’re less worth paying attention to on average.

In this particular case… when you talk about “We don’t want you to say that” comments, it sounds to me like those comments don’t want you to say your ideas. It sounds like Habryka and other commenters interpreted it that way too.

But my read of the the comment you’re talking about here isn’t that it’s opposed to your ideas. Rather, it doesn’t want you to use a particular style of argument, and I agree with it, and I endorse “we don’t want bad arguments on LW”. I downvoted that post of yours because it seemed to be arguing poorly. It’s possible I missed something; I admittedly didn’t do a close read, because while I’ve enjoyed a lot of your posts, I don’t have you flagged in my brain as “if lsusr seems to be making a clear mistake, it’s worth looking closely to see if the error is my own”.

(I am sad that the “avoid paying your taxes” post got downvoted. It does seem to me like an example of the thing you’re talking about here, and I upvoted it myself.)

philh Dec 9, 2023, 1:04 PM
12 points
0
on: New LessWrong feature: Dialogue Matching
My stance towards almost everyone is “if you want to have a dialogue with me and have a topic in mind, great, I’m probably interested; but I have nothing specific in mind that I want to dialogue with you about”. Which makes me not want to check the box, but some people have apparently checked it for me. I’d appreciate having two levels of interest, along the lines of “not interested/interested/enthusiastic” such that a match needs either interested/enthusiastic or enthusiastic/enthusiastic.

philh Dec 5, 2023, 4:41 PM
2 points
0
on: Process Substitution Without Shell?
I’m curious how bash does it. Suppose you do cat <(cat), this gives you
- One cat process with stdout writing to a pipe
- One cat /proc/self/fd/11 process with file descriptors 3 and 11 both reading from this pipe.
So if I had to guess: bash calls pipe() to get two ends of a pipe. Let’s say these have file descriptors 10 and 11. Then it forks twice, which copies the list of file descriptors. We now have three processes with file descriptors 10 (open for writing) and 11 (open for reading). Parent process closes both ends, one child closes the read end, one child closes the write end.

The writing child process uses dup2() or something to set its stdout to be the pipe, then calls exec("cat", []) or whatever.

The reading child process has open file descriptor 11, and can access it as a file at /proc/self/fd/11. So it calls exec("cat", ["/proc/self/fd/11"]). The process which is now cat still has that open file descriptor, but it doesn’t know that, it just knows it’s been given a path; it opens that path and gets file descriptor 3, also reading from that pipe.

So if you want to duplicate this in python:
- You’ll need to figure out how to access the pipe() function.
- Do the various subprocess libraries close file descriptors? If not, you can maybe just pass them /proc/self/fd/(fileno) as the path to read/write. If they do, you might need to do the fork and exec manually?
- Maybe you also need to figure out “use this specific file descriptor for the stdin/stdout of this process”? Not sure. At minimum a fork/exec would work if you do need to do this, but maybe there’s something better.
None of this is very high confidence.

philh Nov 11, 2023, 1:12 PM
2 points
0
in reply to: AlexMennen’s comment on: When and why should you use the Kelly criterion?
You could argue that the latter is more important, since getting high expected utility in the end is the whole point. But on the other hand, when trying to decide on a bet size in practice, there’s a limit to the precision with which it is possible to measure your edge, so the difference between optimal bet and Kelly bet could be small compared to errors in your ability to determine the Kelly bet size, in which case thinking about how optimal betting differs from Kelly betting might not be useful compared to trying to better estimate the Kelly bet.
So like, this seems plausible to me, but… yeah, I really do want to distinguish between
- This maximizes expected utility
- This doesn’t maximize expected utility, but here are some heuristics that suggest maybe that doesn’t matter so much in practice
If it doesn’t seem important to you to distinguish these, then that’s a different kind of conversation than us disagreeing about the math, but here are some reasons I want to distingish them:
- I think lots of people are confused about Kelly, and speaking precisely seems more likely to help than hurt.
- I think “get the exact answer in spherical cow cases” is good practice, even if spherical cow cases never come up. “Here’s the exact answer in the simple case, and here are some considerations that mean it won’t be right in practice” seems better than “here’s an approximate answer in the simple case, and here are some considerations that mean it won’t be right in practice”.
  - Sometimes it’s not worth figuring out the exact answer, but like. I haven’t yet tried to calculate the utility-maximizing bet for those other utility functions. I haven’t checked how much Kelly loses relative to them under what conditions. Have you? It seems like this is something we should at least try to calculate before going “eh, Kelly is probably fine”.
- I’ve spent parts of this conversation confused about whether we disagree about the math or not. If you had reliably been making the distinction I want to make, I think that would have helped. If I had reliably not made that distinction, I think we just wouldn’t have talked about the math and we still wouldn’t know if we agreed or not. That seems like a worse outcome to me.
Why specifically higher? You must be making some assumptions on the utility function that you haven’t mentioned.
Well, we’ve established the utility-maximizing bet gives different expected utility from the Kelly bet, right? So it must give higher expected utility or it wouldn’t be utility-maximizing.

philh Nov 10, 2023, 9:36 AM
2 points
5
on: How to (hopefully ethically) make money off of AGI
And to take it one step further, holding long term debt at fixed rates is amazing in that situation, such as a long term mortgage.
(This is a typo that reverses the meaning, right? Should be “owing” long term debt, you want to owe a mortgage rather than to have issued a mortgage.)

philh Nov 8, 2023, 12:46 PM
4 points
0
in reply to: AlexMennen’s comment on: When and why should you use the Kelly criterion?
Thanks for clarifying! Um, but to clarify a bit further, here are three claims one could make about these examples:
1. As $wealth \to \infty$ , the utility maximizing bet at given wealth will converge to the Kelly bet at that wealth. I basically buy this.
2. As $n \to \infty$ , the expected utility from utility-maximizing bets at timestep $n$ converges to that from Kelly bets at timestep $n$ . I’m unsure about this.
3. For some finite $n$ , the expected utility at timestep $n$ from utility-maximizing bets is no higher than that from Kelly bets. I think this is false. (In the positive: I think that for all finite $n$ , the expected utility at timestep $n$ from utility-maximizing bets is higher than that from Kelly bets. I think this is the case even if the difference converges to 0, which I’m not sure it does.)
I think you’re saying (2)? But the difference between that and (3) seems important to me. Like, it still seems that to a (non-log-money) utility maximizer, the Kelly bet is strictly worse than the bet which maximizes their utility at any given timestep. So why would they bet Kelly?
Here’s why I’m unsure about 2. Suppose we both have log-money utility, I start with $2 and you start with $1, and we place the same number of bets, always utility-maximizing. After any number of bets, my expected wealth will always be 2x yours, so my expected utility will always be $log (2)$ more than yours. So it seems to me that “starting with more money” leads to “having more log-money in expectation forever”.
Then it similarly seems to me that if I get to place a bet before you enter the game, and from then on our number of bets is equal, my expected utility will be forever higher than yours by the expected utility gain of that one bet.
Or, if we get the same number of bets, but my first bet is utility maximizing and yours is not, but after that we both place the utility-maximizing bet; then I think my expected utility will still be forever higher than yours. And the same for if you make bets that aren’t utility-maximizing, but which converge to the utility-maximizing bet.
And if this is the case for log-money utility, I’d expect it to also be the case for many other utility functions.
...but something about this feels weird, especially with $log (money + x)$ , so I’m not sure. I think I’d need to actually work this out.
Here’s a separate thing I’m now unsure about. (Thanks for helping bring it to light!) In my terminology from on Kelly and altruism, making a finite number of suboptimal bets doesn’t change how rank-optimal your strategy is. In Kelly’s terminology from his original paper, I think it won’t change your growth rate.
And I less-confidently think the same is true of “making suboptimal bets all the time, but the bets converge to the optimal bet”.
But if that’s true… what actually makes those bets suboptimal, in those two frameworks? If Kelly’s justification for the Kelly bet is that it maximizes your growth rate, but there are other bet sizes that do the same, why prefer the Kelly bet over them? If my justification for the Kelly bet (when I endorse using it) is that it’s impossible to be more rank-optimal than it, why prefer the Kelly bet over other things that are equally rank-optimal?

philh Nov 8, 2023, 8:32 AM
4 points
0
in reply to: Minetta’s comment on: When and why should you use the Kelly criterion?
Oh, right! I would have been able to give that proof ten years ago, but I’ve forgotten a lot since leaving university.

philh Nov 7, 2023, 7:23 AM
2 points
0
in reply to: AlexMennen’s comment on: When and why should you use the Kelly criterion?
Okay, “Kelly is close to optimal for lots of utility functions” seems entirely plausible to me. I do want to note though that this is different from “actually optimal”, which is what I took you to be saying.

(The thing you actually wrote is the same as log utility, so I substituted what you may have meant)

Oops! I actually was just writing things without thinking much and didn’t realize it was the same.

philh Nov 6, 2023, 4:03 PM
2 points
0
in reply to: Olli Järviniemi’s comment on: When and why should you use the Kelly criterion?
Yes, but there’s an additional thing I’d point out here, which is that at any finite timestep, Kelly does not dominate. There’s always a non-zero probability that you’ve lost every bet so far.
When you extend the limit to infinity, you run into the problem “probability zero events can’t necessarily be discounted” (though in some situations it’s fine to), which is the one you point out; but you also run into the problem “the limit of the probability distributions given by Kelly betting is not itself a probability distribution”.

philh Nov 6, 2023, 1:30 PM
1 point
−3
in reply to: RationalDino’s comment on: When and why should you use the Kelly criterion?
(Variance is “expected squared difference between observation and its prior expected value”, i.e. variance as a concept is closely linked to the mean and not so closely linked to the median or mode. So if you’re talking about “average” and “variance” and the average you’re talking about isn’t the mean, I think at best you’re being very confusing, and possibly you’re doing something mathematically wrong.)

philh Nov 6, 2023, 10:25 AM
2 points
0
in reply to: AlexMennen’s comment on: When and why should you use the Kelly criterion?
So I claim that Kelly won’t maximize $E (\sqrt{money})$ , or more generally $E ({money}^{x})$ for any $x$ , or $E (1 - e^{- money})$ , or $E (log (\sqrt{money}))$ , or $E (min (money, x))$ , or even $E (log (money + x))$ but it’ll get asymptotically close when $money >> x$ . Do you disagree?
Your “When to act like your utility is logarithmic” section sounds reasonable to me. Like, it sounds like the sort of thing one could end up with if one takes a formal proof of something and then tries to explain in English the intuitions behind the proof. Nothing in it jumps out at me as a mistake. Nevertheless, I think it must be mistaken somewhere, and it’s hard to say where without any equations.