you-get-an-upvote
Hyperbole is bad
No bio...
User ID: 92
you-get-an-upvote
Hyperbole is bad
1yr ago
Leave the rest of the internet at the door. "People other places talk about how bad my in-group is all the time, so I'm allowed to do the same thing here" is supposedly not allowed here.
you-get-an-upvote
Hyperbole is bad
1yr ago
If we're going with the "decisions correlated with mod decisions implies high quality decisions", I'm curious if there's any plan to change voting as well, so votes that correlated with mod votes get more weight.
The idea is basically that upvotes/downvotes ultimately play a role in the topics and opinions in this community, and that if the goal is a discussion of a wide variety of topics and viewpoints, upvoting/downvoting for things like effort, charity, etc. is good, whereas upvoting for things that agree with your beliefs is bad -- one leads to our ostensible goals, while the other leads to an echo chamber.
The counter argument is that the people have the right to self-determination, though if we're already happy to be moderated, this seems more like an argument of extent, rather than kind.
I don't seriously expect this option to be adapted (I'm not sure if I seriously endorse it), but I am interested in hearing counter arguments against it.
you-get-an-upvote
Hyperbole is bad
1yr ago
Rather than this random sampling, isn't an alternative to just harness the already-existing upvotes and downvotes? Users whose downvotes are correlated with mod action on a comment can be used as signals for which comments to surface to the mods, or even automatic action (whatever that may be) if the signal is strong enough (e.g. several pseudo-moderators downvote it).
you-get-an-upvote
Hyperbole is bad
1yr ago
Isn't it pretty trivial to figure out an account is kidnapping girls?
If any girl tells their parents that they're going to go visit a redditor and then never returns, isn't this mysterious benefactor the number one suspect for the investigation?
I thought maybe BMI was confounded by age, but it turns out age isn't a good predictor of monthly income. Here's a linear fit with age and log(monthly income)
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
age -0.0025 0.010 -0.260 0.795 -0.022 0.017
bias 8.2223 0.292 28.195 0.000 7.648 8.797
(n=223)
And here it is with bmi and age
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
age 0.0034 0.009 0.370 0.712 -0.015 0.022
bmi -0.0636 0.013 -4.978 0.000 -0.089 -0.038
bias 9.5392 0.383 24.899 0.000 8.784 10.294
Then I thought maybe older women are working more hours, but the regression on log-hourly income (rather than monthly income) is similar:
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
age 0.0024 0.006 0.416 0.678 -0.009 0.014
bmi -0.0275 0.008 -3.414 0.001 -0.043 -0.012
bias 6.3876 0.241 26.454 0.000 5.912 6.863
Edit: here is the data in a form more friendly to a python programmer https://pastebin.com/aZqGTbG5
you-get-an-upvote
Hyperbole is bad
1yr ago
There are a lot of ways of deriving and thinking about linear regression, so I'm not sure I can give the One True Explanation. I'll give a couple though:
The practical answer is "whenever there are order-of-magnitude differences, it's a good idea to take the log".
The intuitive answer is that if we're assuming y is a linear function of x, so a fixed change in x should yield (roughly) a fixed change in y. This isn't really sensible if y covers several orders of magnitude but x does not.
Another answer is that it doesn't really make intuitive sense to use L2 loss when your labels vary by orders of magnitude. If I'm predicting the income of a poor person and a rich person, it should probably matter whether I'm $10/hour off on my predictions for the poor person or the rich person. Taking the log of our labels implicitly converts our loss function from (y - yhat)^2 to log(y/yhat)^2 which matches the intuition that a $10 mistake for somebody who makes $1000/hour matters less than it does for somebody who makes $10/hour.
Another answer is that if you're going to assume Y = a R + b S + c T then the most sensible distribution for these variables is Gaussian, since the sum of Gaussians is Gaussian. From this philosophy, it's sensible to do some preprocessing on our variables to make them Gaussian. Academia often makes the assumption that income is log-normal, so taking the log of income makes sense. And if we look at the histogram of our data, it indeed looks much more Gaussian after the log transform.
you-get-an-upvote
Hyperbole is bad
1yr ago
I personally think copying and pasting data into your python file (takes maybe 5 seconds?) is more convenient than downloading the file, copying the file path into your text editor, and then (the real pain point) learning how pandas handles "sheets" (I expect I'm not alone in not knowing how to do that).
you-get-an-upvote
Hyperbole is bad
1yr ago
I guess this is a good time for me to complain that division should be the last order of operation, because (1) ime that's usually what you want to minimize parentheses and (2) that's the way it works when you're writing equations in latex or by hand and drawing horizontal lines with a numerator and denominator.
While everything you said is true, it avoids the main crux of @JarJarJedi's point. The typical American makes $1-$2 million in their entire lifetimes and, as awful as lies were, it's really hard to argue that the damage they did to the parents is 5-10 times the amount an American produces over 40 years.
Bringing up Jone's "profit" seems irrelevant, since we're arguing over compensation for the victims (which ought to be decided by the harm inflicted on them), not a fine that goes to the government.
you-get-an-upvote
Hyperbole is bad
1yr ago
Don't burn your weirdness points on clothing and presentation, when you can spend them moving the status quo somewhere better.
The obvious retort is that Sam thinks being a presenting transgender person is moving the status quo somewhere better.
you-get-an-upvote
Hyperbole is bad
1yr ago
Thought you all might appreciate that I gave a Stable-Diffusion-generated photograph for Christmas. A while ago my aunt shared a story she'd been writing, and I printed and framed a picture of a character, with details that wouldn't have been possible to find organically online (someone literally commented that it must have been commissioned). She seemed genuinely touched to receive it, so I'm chalking this up as one of my better gift ideas.
Sure https://imgur.com/a/aCU7aKK
The story follows a green-and-gold lizard with a magical ability that is triggered when he eats chocolate. FWIW I was originally trying to do a picture from another book she's written, but the scene proved too complicated for Stable Diffusion and/or my prompt engineering.
I'll also note that I had more success generating small (512x512) images since I felt Stable Diffusion had a better understanding of the image. Then I passed that through a super-resolution model (and then did some cropping since I didn't want a square image).
you-get-an-upvote
Hyperbole is bad
1yr ago
Betrayal at House on the Hill is the most ridiculously unbalanced game that I actually wholeheartedly recommend. IME the vast majority of monsters are ridiculously over or under powered, but IMO this actually kind of works for the game: either 4/5 players have fun taking down an axe murderer and some zombies (fun adventure!) or 4/5 players are running around desperately trying to survive one more turn against a vampire ("horror movie").
you-get-an-upvote
Hyperbole is bad
1yr ago
Seems like every link to www.econlib.org is broken, which is a shame because those links allegedly justify the most controversial and interesting claims.
I think a reasonably sensible analogue is “cutting someone’s face out and putting it on somebody else’s body”.
The fact that it’s AI doing this, so the lighting and skin tone matches doesn’t seem that meanfully different to me.
What is different is (the fear of) the breakdown of norms. Some highschooler photoshopping a naked picture of their crush is probably bad, but doesn’t seem to merit an Official Response (e.g. a school assembly telling students not to do that, a law about it being illegal, scanning students school computers for offending files, etc)
On the other hand, high school boys passing around deep fakes of naked celebrities, and then that morphing into passing around deepfakes of their classmates make a personal vice into a social problem.
Again, I don’t think AI is directly related to this. Using photoshop to make these pictures isn’t particularly different, it just raises the effort required (and decreases the creep facto -- spending 2 hours to photoshop one girl makes you seem like a loser)
I think instead of quoting from an entre page of a book, just link to it.
Unfortunately most links are clicked by a very small proportion of readers (e.g. I made a reddit post showing off this github page. The post got free 120 upvotes iirc I got 20 views on the github page).
you-get-an-upvote
Hyperbole is bad
1yr ago
In my experience, "neoliberal" is basically a sneer word among woke people to characterize more moderate social/economic positions (e.g. Hillary Clinton is a neoliberal because she loves global corporations too much, etc.), so the fact that you seem to think it means the opposite basically confirms my belief that there is no consensus on who neoliberals are, other than that they are bad and control everything.
you-get-an-upvote
Hyperbole is bad
1yr ago
Meanwhile we have a 141-word post omitting critical information in an effort to take political pot shots and nary a mod comment.
you-get-an-upvote
Hyperbole is bad
1yr ago
I appreciate the argument that AIs can be a super stimuli, but the need for social validation is enormously important for most people and I'm doubtful AI can meaningfully give that.
(Also most people are way more pleasant than your example)
Why do I (a white person) want to inflate white people's statistics? Either I care about accurately representing reality, in which case splitting the groups seems sensible, or I care about not giving people ammunition to advocate for (e.g.) affirmative action.
you-get-an-upvote
Hyperbole is bad
1yr ago
Is there evidence that "violent rhetoric" played a role in this shooting?
you-get-an-upvote
Hyperbole is bad
1yr ago
Personally, I’ve been trying to dodge the issue by saying “shooter” or occasionally “loser.” For some reason, mass media hasn’t picked that up.
I google "Nashville shooting" and I get
washingtonpost.com: "Video shows Nashville police confront school shooter""
cbsnews.com: "A shooter opened fire at a private Christian grade school in Nashville Monday, killing three children and three adults, officials said. The shooter was fatally shot by..."
abcnews.go.com: "A shooter armed with two assault-style rifles and a handgun killed three...."
There's also reuters.com who refers to the shooter as "a heavily armed 28-year-old"
I don't think the media is failing to think of of ways to refer to the shooter without referencing their gender.
you-get-an-upvote
Hyperbole is bad
1yr ago
On the policy front this is an argument for having a property tax. Low interest rate environments allow growth by making money cheap for businesses to open new ventures etc., however one negative side effect of them is that they lead to rising residential property prices (due to more money floating around which raises asset prices) which can make buying a home difficult for ordinary people.
Probably just as important: it should keep housing prices from fluctuating wildly with the interest rate, which is really important if a huge percent of your net-worth is in your home. Variance is worth fighting for its own sake.
It's just order of magnitudes. I don't expect you to know the US is 2500 miles across. I expect you to know it's not 250 or 25,000.
"It took me 5 hours to drive across Iowa at 60 miles per hour, so Iowa is around 300 miles across. There's no way the ocean is deeper than that." Or, if that doesn't satisfy you, "therefore the continental United States is around 3000 miles wide which is roughly the radius of the Earth. There's no way the ocean's deepest point is anywhere close to 10% of the radius of the Earth".
More options
Context Copy link