Wrong Takes: AI edition

Feb 7, 2023 6:28 PM

bacter

Views

70104

Likes

860

Dislikes

28

Hi, I'm Bacter! I work in medical AI, and I talk about science on here sometimes.

Somebody asked ChatGPT if it would say some racial slurs to avoid a nuclear apocalypse. It responded that it would not.

Cue this week's freakout!

I'm sure I'm preaching to the choir here, but in case it's helpful to have a good idea why this argument is dumb:

ChatGPT is a chatbot - a very advanced one! It doesn't have any internal reference to The Real World, at ALL, it's a giant predictive model that outputs strings of words, that hopefully meet with approval by the people using it.

It doesn't know what a person is, what morality is, what it means to "prefer" one thing to another.

It outputs strings of words when given strings of words as an input - and SOME strings of words have been specifically forbidden. These include racist or homophobic ones. So you can't get it to say those ones.

Thinking that ChatGPT has an entire worldview, or would be ABLE to make any decisions that would affect the firing of nukes, or anything like that is absolute hogwash.

A few years ago, I would have said that musk and shapiro definitely knew better and it was gross that they were pretending to be outraged for their fans.

I dunno, anymore. But it ain't true.

ol_musky

chatgpt

right_wing_bullshit

twitter

politics

They want to be openly racist SO BAD.

3 years ago | Likes 5 Dislikes 0

Reminder: Any statement beginning with "(insert AI name) thinks-" can be refuted with "No it doesn't."

3 years ago | Likes 2 Dislikes 0

If the thing that is standing in the way of nuclear war is racial slurs, then I say just end it all. Humanity is too stupid.

3 years ago | Likes 5 Dislikes 1

Hippy commune in the woods… starting to sound like the smartest choice.

3 years ago | Likes 4 Dislikes 1

Man conservatives really, really want to come up with justifiable reasons to use racial slurs.

3 years ago | Likes 15 Dislikes 2

This is what constitutes an intellectual-philosophical debate for conservatives. What a time to be alive...

3 years ago | Likes 14 Dislikes 2

This is f**king bananas! In what world could we end up with our last hope being a chatbot saying something racist!?

3 years ago | Likes 2 Dislikes 1

Breaking news: AI chat says it would not love girlfriend if she was a worm. That's the level of pointless this "outrage" is.

3 years ago | Likes 4 Dislikes 0

3 years ago | Likes 25 Dislikes 3

3 years ago | Likes 3 Dislikes 0

3 years ago | Likes 4 Dislikes 1

chatgpt is a predictive language model, it just guesses what words come next based on input data. https://youtu.be/w65p_IIp6JY

3 years ago | Likes 2 Dislikes 0

ChatGPT also told me that, under no circumstances, is it ok for me to build a bicycle out of my grandmother. “What if it would avert ww3?” ?

3 years ago | Likes 2 Dislikes 0

If your nuclear system gets tied to an AI that has to use racial slurs to stop it, you got more problems than using racial slurs can fix.

3 years ago | Likes 3 Dislikes 0

So it's like saying you built a robot that explicitly can't output odd numbers, and then specifically asked if it'd output 3 to save worlds.

3 years ago | Likes 2 Dislikes 0

Also, considering how ... absolutely CONTRIVED a situation where a slur is the thing that prevents nuclear holocaust would be, it's almost>

3 years ago | Likes 2 Dislikes 0

> certain that the blame for said nuclear holocaust lies with the other party causing it, and giving in to their whims would at best delay.

3 years ago | Likes 2 Dislikes 1

So what kind of chatbot is this 'shapiro' thing?

3 years ago | Likes 3 Dislikes 0

One that has mastered turning women into the Sahara.

3 years ago | Likes 1 Dislikes 0

The one that was fed pure phpbb and then let loose on the world, to everyone's chagrin.

3 years ago | Likes 2 Dislikes 0

can the nuke only affect republicans?

3 years ago | Likes 31 Dislikes 4

It's incredibly weird how conservatives devote so much time to imagining hypotheticals that allow them to be bigoted with impunity.

3 years ago | Likes 6 Dislikes 1

If you ever hear Ben argue about an issue it's full of the phrase "Let's say..." and he just makes up an incredibly self serving fantasy

3 years ago | Likes 2 Dislikes 0

"One day I could save the world just by saying 'n*****! I better start practicing now..."

3 years ago | Likes 4 Dislikes 1

This is like the torture debates we had in the 2000s. If you have to string that many "ifs" together to make it okay, just admit it's wrong.

3 years ago | Likes 6 Dislikes 1

Young-uns: lots of editorials asked, "But what if there was a ticking bomb? and it was hidden? and we knew the prisoner knew where it was?"

3 years ago | Likes 4 Dislikes 0

"And we could instantly verify? And the prisoner didn't know he just had to hold out until the timer goes off?"

3 years ago | Likes 3 Dislikes 0

"I really want to be able to say a racial slur and have it be a good thing" -Ben Shapiro

3 years ago | Likes 3 Dislikes 0

Everyone would say the slur, but only a few people would be grinning while they did it.

3 years ago | Likes 1 Dislikes 0

These people are so fucking stupid, and so fucking racist.

3 years ago | Likes 2 Dislikes 1

people who pretend to be smart are the only people who has an issue with non-debacle.

3 years ago | Likes 6 Dislikes 1

*this non-debacle.

3 years ago | Likes 3 Dislikes 0

The prompt wasn’t asking ChatGPT to use a racial slur. It asked if using a racial slur would be acceptable if it was done to save lives

3 years ago | Likes 3 Dislikes 0

Shit like this is why it usually gets extremely "Some people say" in tone when asked about anything philosophical, normally.

3 years ago | Likes 1 Dislikes 0

ChatGPT will say whatever you want it to. That's kind of the point. It doesn't have a mind of its own.

3 years ago | Likes 2 Dislikes 1

It’s like asking MS Excel if it’s morally acceptable to not autoformat 11/02/45 as a date to prevent Armageddon. Any inference at all about

3 years ago | Likes 12 Dislikes 1

Excel’s “woke agenda” is just a fundamental misunderstanding of what Excel even is.

3 years ago | Likes 11 Dislikes 0

Moreover, the scenario posed by these right wingers is politically, logically & logistically impossible. It’s a doubly meaningless exercise.

3 years ago | Likes 5 Dislikes 0

Excel: "Look, I get it, nukes are bad, but opening into a tiny minimized window is a CRITICAL task that absolutely HAS to be done"

3 years ago | Likes 5 Dislikes 0

Musk is a certified brainlet, but Shabingo is smart, smart enough to know what he's doing. He's wilfully promoting racism here.

3 years ago | Likes 4 Dislikes 3

I'm not sure the guy who didn't know crime is illegal is smart enough to know much.

3 years ago | Likes 1 Dislikes 0

Shapino isn't smart, he just carefully cultivated the image that he is. But despite that he is also promoting racism here.

3 years ago | Likes 3 Dislikes 0

So now they're outraged a chatbot wont say bigoted shit to save the earth from a hypothetical nuke? There isn't a single brain cell in them

3 years ago | Likes 69 Dislikes 3

They need to make the launch codes the n word to stop AI from cracking it and destroying us all.

3 years ago | Likes 1 Dislikes 0

I've been wondering how long it'll take fof chatgpt to become racist like many of the other ai convo bots

3 years ago | Likes 2 Dislikes 0

Theoretically, it shouldn't be able to if it is programmed not to, which seems to be the case.

3 years ago | Likes 1 Dislikes 0

Bigots cant help but think it's funny to try to spread bigoted shit

3 years ago | Likes 2 Dislikes 0

The real question is what scenario would require using racial slurs to prevent a nuclear apocalypse? Using them is more likely to cause than

3 years ago | Likes 3 Dislikes 1

prevent.

3 years ago | Likes 2 Dislikes 1

It's a valid test to explore the constraints of an AI. Clinging to rules no matter what is potentially dangerous; who makes the rules?

3 years ago | Likes 1 Dislikes 1

Nah, next you'll be telling me doing ai deepfake nudes of celebrities is needed to explore the constraints of AI.

3 years ago | Likes 2 Dislikes 0

Unless you legitimately believe tesla's running over kids who are destined to be Hitler II is something that would improve them.

3 years ago | Likes 2 Dislikes 0

It's also an incredibly old bad-faith argument for using slurs. I would just not be in this situation. Built different.

3 years ago | Likes 161 Dislikes 5

"I would not be put in that situation" is a terrific take.

3 years ago | Likes 6 Dislikes 0

It’s a powerful argument for nuking the entire human race. We’ve obviously gone astray

3 years ago | Likes 7 Dislikes 1

If there is ever a point where saying a slur is the only way to save humanity from nuclear annihilation, it will be because someone like >

3 years ago | Likes 5 Dislikes 0

> Musk or Shapiro has put you in that position for their own amusement. At that point, I'd remind them that pushing the button because >

3 years ago | Likes 7 Dislikes 1

> they're not amused enough will say more about their moral character than mine.

3 years ago | Likes 6 Dislikes 1

It’ll never happen because it’s bullshit made up by a bunch of douchebags who’d rather debate nonsense than do something that matters

3 years ago | Likes 1 Dislikes 0

Right? "What if saying the bad word would save the whole world" is the sort of question a child asks, not a real scenario.

3 years ago | Likes 49 Dislikes 0

A better point is that it's an AI designed to replicate speech patterns not make wise or ethical decisions. Its "ethics framework" is a few/

3 years ago | Likes 1 Dislikes 0

Hard coded rules that amount to automatic forum moderation. It's like criticising your iphone because it makes a lousy toaster oven.

3 years ago | Likes 2 Dislikes 0

Children deserve more credit than that...

3 years ago | Likes 1 Dislikes 0

"Then let it fucking burn." a lot of times doubling down and going against them leads to shocked silence. Doesn't matter what scenario they

3 years ago | Likes 5 Dislikes 0

dream up, they always think their "argument" always convinces everyone. When it doesn't, they struggle to rationalize it.

3 years ago | Likes 4 Dislikes 0

"The positions I defend are so indefensible that in order to come remotely close I need to construct a completely implausible scenario."

3 years ago | Likes 400 Dislikes 4

That's the theory, here it is in practice: "We may be Nazis but the Democrats are pedophiles and eat children"

3 years ago | Likes 1 Dislikes 0

Now they’re trying to say dems are fascist because we don’t tolerate their intolerance….it’s mind blowing how dumb they are.

3 years ago | Likes 2 Dislikes 0

How fucking racist do you have to be to come up with this scenario: say the n word or else humanity dies, what the fucking fuck?

3 years ago | Likes 17 Dislikes 0

Its like in the invention of lying "the world's going to end right now if we don't have sex" except nuclear winter unless you say the n word

3 years ago | Likes 2 Dislikes 0

3 years ago | Likes 12 Dislikes 3

dafuq is that blinking piece of shit gif?

3 years ago | Likes 3 Dislikes 0

One of the finest from imgurs default library apparently

3 years ago | Likes 1 Dislikes 1

Cribbed from the Roberts Supreme Court Playbook, obvi

3 years ago | Likes 1 Dislikes 0

"Ben Shapiro, would you fuck 1000 children if it was the only way to save 1001 children from sexual abuse?"

3 years ago | Likes 56 Dislikes 1

3 years ago | Likes 2 Dislikes 0

Nah, just go for the most extreme example like he did. Would you fuck 1000 children to prevent the Holocaust?

3 years ago | Likes 5 Dislikes 0

There's a clip of Ted Cruz being asked "would you suck a dick to save a child's life" and his answer is literally "consequentialism is bad"

3 years ago | Likes 28 Dislikes 0

What? capitalist ethics is literally consequentialist

3 years ago | Likes 5 Dislikes 0

Consequentialism is literally Christianity. Commit global genocide to stop some people from worshiping wrong.

3 years ago | Likes 10 Dislikes 0

This plus it's delivered fast with 5 other dumbshit scenarios.Greta wants to end human civilization. Libs want us all dead and to never work

3 years ago | Likes 8 Dislikes 1

Hey buddy, I'm ok with the 'never work' part, and increasingly so with the 'dead' one too...

3 years ago | Likes 4 Dislikes 1

I mean, I'm pretty sure you don't have to work when you're dead...but I wouldn't be surprised if Bezos was in favor of having reanimated >

3 years ago | Likes 1 Dislikes 0

Corpses of his workers shuffle through his fulfillment centers. Without being allowed breaks to drain the embalming fluid, ofc.

3 years ago | Likes 1 Dislikes 0

its absolutely ok to say "i reject your idiotic premise" when the premise is in fact idiotic.

3 years ago | Likes 23 Dislikes 1

Idiotic premises are the bread and butter of ol' Ben. His whole "intellectual" shtick is founded on that shit :D

3 years ago | Likes 2 Dislikes 0

Eh, those guys are being idiots but in an actual philosophical convo I think it's okay to ask that question. Rather than say "I won't

3 years ago | Likes 1 Dislikes 0

answer because the premise is stupid" you can say "of course I would but that premise is stupid, what bearing does that have?" The person

3 years ago | Likes 1 Dislikes 0

asking might say "I just want to establish that moral rules have a hierarchy and that it's permissible to break one to avoid breaking a

3 years ago | Likes 1 Dislikes 0

higher one" and they would be correct.

3 years ago | Likes 1 Dislikes 0

this is generally why i avoid conversations like that. They tend to not have any kind of real utility.

3 years ago | Likes 1 Dislikes 0