|
ultrafilter posted:The actual abstract isn't as bad as a lot of what people think the paper might be about.
|
# ? Sep 25, 2020 17:07 |
|
|
# ? May 13, 2024 07:34 |
|
qirex posted:also classical portraits seems like the worst possible training set since they were usually done specifically to make the patron look awesome and their enemies look shady yeah, capital p Portraits arent naturalistic, not then, not now. they all have a message. i guess theres value in examining what that message is and how it has changed through history, but thats not a job for machine learning lol
|
# ? Sep 25, 2020 17:09 |
|
ultrafilter posted:If I were an art historian I'd probably have a lot to say about this but as is I'm just withholding judgment until the actual experts weigh in. sounds like theyre evaluating historical and contemporary portraits based on whatever the current model of "trustworthiness" says, ie they show that the common styles of portraits indeed change from historical styles towards the contemporary styles, and that the current model considers historical portraits less trustworthy, whatever that menas im tired
|
# ? Sep 25, 2020 17:13 |
|
“let’s use a set of images created entirely for the purpose of flattering rich white people to train for trustworthiness” - a “scientist”
|
# ? Sep 25, 2020 17:21 |
|
https://twitter.com/baumard_nicolas/status/1308715649766699010?s=20
|
# ? Sep 25, 2020 18:52 |
|
that tweet took a turn
|
# ? Sep 25, 2020 19:30 |
|
I retract my previous statement about it being a bad training set now that the author has clarified that the did it specifically to discriminate against poor people
|
# ? Sep 25, 2020 19:37 |
|
does “have lower levels of trust” mean “they should be trusted less” or “they should trust others less” there? i read the vulnerable to exploitation part of it to imply the latter.
|
# ? Sep 25, 2020 19:49 |
|
Midjack posted:does “have lower levels of trust” mean “they should be trusted less” or “they should trust others less” there? i read the vulnerable to exploitation part of it to imply the latter. the phrasing "trustworthiness" implies the former
|
# ? Sep 25, 2020 19:52 |
|
Carthag Tuek posted:the phrasing "trustworthiness" implies the former read the article, bottom of page 2. the argument on this statement is a little convoluted, the author is saying that as gdp goes up general societal trust goes up and he’s using trustworthiness ratings of portraits as a proxy for general societal trust. edit: yeah boy howdy this is kind of junk science and i thought nature was better than this. Midjack fucked around with this message at 20:03 on Sep 25, 2020 |
# ? Sep 25, 2020 20:00 |
|
https://twitter.com/JedMSP/status/1309525866746056709
|
# ? Sep 25, 2020 20:01 |
|
Midjack posted:read the article, bottom of page 2. the argument on this statement is a little convoluted, the author is saying that as gdp goes up general societal trust goes up and he’s using trustworthiness ratings of portraits as a proxy for general societal trust. Is this some sort of weird p-hacking thing like where the cup size of Miss America is inversely correlated to the unemployment rate in Wyoming or something? Like, in what world are "portraits being smiley or frowny" even a variable?
|
# ? Sep 25, 2020 20:05 |
|
ultrafilter posted:If I were an art historian I'd probably have a lot to say about this but as is I'm just withholding judgment until the actual experts weigh in. im not. this seems terrible what even is their ground truth value for "trust"? did they have a grad student go through portraits and be like "eh, that one looks shifty, I'll give em a three"
|
# ? Sep 25, 2020 20:12 |
|
Midjack posted:read the article, bottom of page 2. the argument on this statement is a little convoluted, the author is saying that as gdp goes up general societal trust goes up and he’s using trustworthiness ratings of portraits as a proxy for general societal trust. Nature Communications charges a $5,700 publishing fee too lol
|
# ? Sep 25, 2020 20:22 |
|
im in awe at the stupidity of this model https://twitter.com/baumard_nicolas/status/1308715661154308103 happy smiling man determined by (points at seagull) to be more trustworthy than dour looking angry man! what happens if they run it with trump doing that smarmy grin he does and joe looking scared and confused
|
# ? Sep 25, 2020 20:46 |
|
apropos, this year's ig nobel prize (ceremony last week) for economics was for this:quote:ECONOMICS PRIZE [UK, POLAND, FRANCE, BRAZIL, CHILE, COLOMBIA, AUSTRALIA, ITALY, NORWAY, ITALY] tldr: if combine those two papers & only trustworthy people on the mouth, youll get rich Carthag Tuek fucked around with this message at 21:09 on Sep 25, 2020 |
# ? Sep 25, 2020 21:05 |
|
he’ll yeah
|
# ? Sep 26, 2020 14:19 |
|
https://twitter.com/astroskylar/status/1309618682176450560
|
# ? Sep 26, 2020 18:56 |
|
jehanne butler was right
|
# ? Sep 26, 2020 20:24 |
|
gotta agree with this
|
# ? Sep 27, 2020 00:17 |
|
Cowards didn't open source their algos, so they can't be tested against classical trustworthiness techniques
|
# ? Sep 28, 2020 02:36 |
|
run the trustworthiness model in reverse. what is the most trustworthy possible face. I want to see it
|
# ? Sep 29, 2020 04:31 |
|
Bloody posted:run the trustworthiness model in reverse. what is the most trustworthy possible face. I want to see it turn off your monitor, friend
|
# ? Sep 29, 2020 04:34 |
|
|
# ? Sep 29, 2020 04:38 |
|
in other news, Florida ran a beta test of a predictive policing algorithm, which resulted in... a bunch of random people being harassed for no reason.quote:A central Florida sheriff built an algorithm meant to predict which people in his jurisdiction were likely to commit a crime in the future.
|
# ? Sep 30, 2020 22:09 |
|
|
# ? Oct 1, 2020 19:30 |
|
you see a graph like that you gotta play the curb your enthusiasm theme over it doot doot doot dadadadahdahdah ditdahditdadah..
|
# ? Oct 1, 2020 19:32 |
|
e: wrong thread
|
# ? Oct 1, 2020 19:35 |
|
https://twitter.com/hardmaru/status/1312058776007008256
|
# ? Oct 3, 2020 03:53 |
|
anfd today congratulations we give that piece of poo poo algorithm the ability to vote we took care and almost wated for double age so it doesnt matter if we hosed it up
|
# ? Oct 3, 2020 04:10 |
|
lmao at least my dogshit ml thesis got the roc on the right side of y=x
|
# ? Oct 3, 2020 06:44 |
|
https://twitter.com/ReinH/status/1313547425563377664
|
# ? Oct 6, 2020 21:37 |
|
laughing at the thought of training for trustworthiness when the machine probably can't even recognize non white people to begin with.
|
# ? Oct 7, 2020 17:52 |
|
the kind of machine learning where you train it on a picture with a style and apply that style to another image can someone tell me what the name of the technique is, or recommend a link or something like https://deepart.io/ I wanna gently caress around with it on my own pc. can I do that? can you train it on many images to try to derive a style from many images? my main purpose is to try to see if it's possible to subtly make my renders look, illustrated. might end up looking like total dog poo poo but ya never know
|
# ? Oct 29, 2020 08:20 |
|
called style transfer you can make illustrative style, tuning the dealios is just not hard if youre not doin anything original
|
# ? Oct 29, 2020 08:28 |
|
i’m not sure what you’re trying to say in that last sentence but surely you’ve seen rodtronics bullshit anyway thanks for the hot tip
|
# ? Oct 29, 2020 10:09 |
|
echinopsis posted:the kind of machine learning where you train it on a picture with a style and apply that style to another image gmic has something like this It can be scripted too last time I used it
|
# ? Oct 29, 2020 10:12 |
|
cool ive found this https://pypi.org/project/neural-style/ and gonna get my head round some of the commands also ill have a look at that gimc thing i only want a subtle effect
|
# ? Oct 29, 2020 10:35 |
|
echinopsis posted:the kind of machine learning where you train it on a picture with a style and apply that style to another image Google colab is perfect for this, you run the python on google hardware for free. Basically most AI stuff like this you can search for “___ colab” and find a working example. https://colab.research.google.com/g...Execution.ipynb
|
# ? Oct 29, 2020 16:59 |
|
|
# ? May 13, 2024 07:34 |
|
hmm thanks my first result turned out absolute dogshit and it’s kinda burned me lol
|
# ? Oct 29, 2020 18:59 |