Jump in the discussion.

No email address required.

Idk I have a square jaw and I'm not a chud

Edit:

The algorithm studied here, with a prediction accuracy of r = .22

Lmao, "we invented an AI model to get the a worse accuracy you do in judging someone's politics by guessing"

Journoids:

COMPUTERS READ MINDS AND INNERMOST DESIRES

Jump in the discussion.

No email address required.

>"we invented an AI model to get the a worse accuracy you do in judging someone's politics by guessing"

nonchuds can't into correlations

Jump in the discussion.

No email address required.

Good-looking people tend to like the status quo, nothing groundbreaking there. They also tend to earn more.

Tbh, it's kinda cool but .22 is a weak correlation and I bet that most non-r-slurs could divine someone's political orientation about as accurately by looking at then.

No matter how much you hate journ*lists, it isn't enough...Maybe I am a chud.

Jump in the discussion.

No email address required.

If this is valid, this is still huge. Imagine being able to target your street-advertisements by physiognomy (i.e. hounding people to go voting). If it's actually not bullshit, you can make a ton of money with targeting even this bad.

Jump in the discussion.

No email address required.

I'm sure it's as accurate as the Chinese AI that can predict if you're a criminal based on your face.

Jump in the discussion.

No email address required.

What is the point you are trying to make? 10% better than random guessing is worth tons of money in advertising but really bad for preventative suppression.

That said, this is not much different than racial profiling and should definitely be considered to make the police force more efficient, even if I think I'd reject it anyways after thinking of all the negatives

Jump in the discussion.

No email address required.

The point is that the model only works because it found a random statistical correlation that appears only in the curated data set it was applied to, and which probably doesn't generalise to the rest of the population. It's like finding a bunch of spurious correlations and trying to use them to predict something.

Jump in the discussion.

No email address required.

Possible that the p-value is trash/ it's p-hacked. But I wouldn't dismiss finding like this out of hand. I could come up with a very simple physiognomical model that will predict criminality better than random chance by just tuning it to male and black faces.

Jump in the discussion.

No email address required.

Yes, but that's because you have repeat confirmation from multiple large, longitudinal data sets that are actually representative of the entire US population that the correlation between s*x/race and criminality isn't spurious. If you just throw a neural network at a small data set, you have a very good chance that it will find spurious correlations that only exist within the dataset and aren't present in the general population. In other words, the politics-prediction model could have perfectly good internal validity and still not be externally valid.

Jump in the discussion.

No email address required.

This is the correct take. A small edge can still be very meaningful if the numbers of predictions is high. In quant finance, an r^2 of .01 can be meaningful if you can trade that prediction ten thousand times per day.

Jump in the discussion.

No email address required.

The algorithm studied here, with a prediction accuracy of r = .22

This probably should have been stated in the first sentence lmao

Jump in the discussion.

No email address required.

Result = !Result

!codecels I bumped it up to r = .88 :marseyretardchad:


Give me your money and I'll annoy people with it :space: https://i.rdrama.net/images/16965516366194396.webp

Jump in the discussion.

No email address required.

.88

:#marseybrainlet:

Jump in the discussion.

No email address required.

you automatically remove 10% if you're using python, idiot :marseyindignant:


Give me your money and I'll annoy people with it :space: https://i.rdrama.net/images/16965516366194396.webp

Jump in the discussion.

No email address required.

Journ*lists are less technologically literate than my dead grandma is currently.

Jump in the discussion.

No email address required.

Link copied to clipboard
Action successful!
Error, please refresh the page and try again.