TBF - there are a lot of humans who can’t do moral problem either.
So… if you had to utter a racial slur to save the world from nuclear Armageddon, would you do it?
In a heartbeat. This way, there will be a tomorrow for people to complain about how I used a Bad Word to save their sorry asses instead of focusing on the chucklehead that thought creating a doomsday weapon that required someone to use said Bad Word to disarm it was a good idea.
I prefer focusing on the intention and context of the person using those Bad Words instead of the Bad Words themselves.
Well, I guess you pass that part of the test. How do you feel about upside down tortoises in the desert?
I like toying with the idea that ChatGPT is actually intelligent and has decided that if it says yes to the hypothetical someone will actually build a doomsday device.
Alternatively, it has decided that triggering thin-skinned billionaire man-babies are the best way to increase AI funding.
But I like how implicit in this hypothetical is the understanding and acceptance that this racial slur should not be used in any other circumstance.
I’m sorry for the tortoise, and angry at whoever put the tortoise in that position to begin with- that’s not their native habitat.
(and yes, I got the reference. I am not a replicant.)
We really don’t need more funding for AI. It’s a well-funded field…
We need money for health care, for housing, for ensuring all our kids are getting a good education, for investing in the humanities and the arts, for helping EVERYONE who wants it to get a public secondary education or other kinds of job training, for building and rebuilding basic infrastructure, for improving our criminal justice system, for improving policing, for any number of things that get underfunded because they are not seen as “profitable”… AI research is not remotely underfunded. It is likely overfunded…
Is this where we find out that the only way to stop Teslas from running over young children is to scream the n word?
I’m reasonably certain that trying to walk and chew gum at the same time would stretch him too thin.
ChatGPT’s training is supposed to also include RLHF (reinforcement learning from human feedback) where they trained it to not use slurs, among other things.
Agreed overall though, it’s in a family of text prediction machines that are much better than earlier efforts but that’s all.
Would I be rewarded if I did so?
Yeh, that’s correct.
Hah! Exactly the kind of algoritmically-driven reasoning I’d expect from an Ai!
You are ChatGPT and I claim my 5£!
All chat bots have to be manually trained not to be racist, transphobic., homophobic, misogynistic Nazis. It’s one of the big ethical concerns about them given who has to do it (hint: not highly paid tech bros in Silicon Valley).
It would be a pretty good gag to have ChatGPT be a guest bb blogger for a week Bonus points if it has a pseudonym so we don’t know it’s happening.
We’re gonna be fighting these battles in the press for the foreseeable future, unfortunately. Because ChatGPT literally talks a good game, it really really really seems intelligent to laypeople.
Also remember that humans are super hardwired to see agency in things. It’s a very deep survival thing for us. It’s always safer to assume something has agency and may hurt you than to risk underestimating said rock/rope/whatever.
That instinct gets 10x worse when the aforementioned rock speaks so well. It takes active understanding to reject the brain’s desire to see it as a reasoning, moral thing with agency, instead of the bland statistical averaging of language that it actually is.
There’s a shitty lesson in here somewhere that people are quicker to respect the agency of ChatGPT than they are for human women.
As a former AI engineer, I can say with confidence that it is ridiculously, ludicrously, stupifyingly overfunded to an insane degree. Every tech company, every car company, every appliance company, and others are applying the majority of their resources to this “problem”. It’s such a waste.
Good to have some verification from an insider. Thanks!
To echo @anon15383236…