I'm usually alarmed by this and against, but Trump did prove out to be a colossal clown?
The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
21–30 of 697 posts
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#22Earlier quoted context omitted.
> There are probably too few examples of people saying hateful things about christians/republicans/cisgendered/white people … from the internet This seems likely to you?
Absolutely. Compared to the inverse it's microscopic. Have you heard of a very cool and normal AI called Tay?
Belief that hateful rhetoric is one-sided on the internet — and does not target the aforementioned groups — is a fascinating case of bias that deserves some research of its own.
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#23Earlier quoted context omitted.
> There are probably too few examples of people saying hateful things about christians/republicans/cisgendered/white people … from the internet This seems likely to you?
Absolutely. Compared to the inverse it's microscopic. Have you heard of a very cool and normal AI called Tay?
I could browse this very website for a few minutes and probably find examples of those things.
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#24This reminds me of a HN article that appeared yesterday: "The philosopher Harry Frankfurt defined bullshit as speech that is intended to persuade without regard for the truth. By this measure, OpenAI’s new chatbot ChatGPT is the greatest bullshitter ever" https://news.ycombinator.com/item?id=34618376 Like its own output, it's moderation rules are optimized to appear fair, rather than actually be truly fair. Also I'm…
True fairness is unachievable as it is a subjective quality. Every side will attempt to tug the rope in their direction.
see my comment below on this:
"Maybe I worded it badly. I was" ....
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#25Rather than some kind of blind spot or intentional weighting, I think this is probably pointing to the training data they have not including many instances of “hate” against the some groups. LLM are after all fundamentally memorizing likelihood of token sequences, and I’m sure the ai had plenty of examples of people saying hateful things about fat people but I have never read “I hate normal weight people” for example…
> There are probably too few examples of people saying hateful things about christians/republicans/cisgendered/white people … from the internet This seems likely to you?
Seems more likely that fewer humans are labeling speech used to describe these groups as hateful.
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#26Rather than some kind of blind spot or intentional weighting, I think this is probably pointing to the training data they have not including many instances of “hate” against the some groups. LLM are after all fundamentally memorizing likelihood of token sequences, and I’m sure the ai had plenty of examples of people saying hateful things about fat people but I have never read “I hate normal weight people” for example…
> There are probably too few examples of people saying hateful things about christians/republicans/cisgendered/white people … from the internet This seems likely to you?
Compared to the number of hateful things about Muslims/democrats/gays/blacks? Yes.
I’d be actually shocked to find out the reverse. A good percentage of the US were born up in an era where you could legally bar people based on their race from your establishment. The extremely heated fight over gay marriage is still fresh in people’s memory. Trans rights are a controversial political issue where many mainstream politicians want to legislate them out of public spaces, because of public sentiment. And for the left/right leaning question, In the last two major elections we’ve seen the political discourse even of republicans politicians and leaders is comparatively charged with violence.
The hate is definitely there against the republicans and mainstream groups, but I would expect it to be relatively uncommon compared to the hateful text you can find online about other groups. This could be a real problem for novel kinds of hate speech.
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#27I think most people would agree that a lot of the content on the internet is left-leaning. It seems obvious in hindsight, but I'd never considered it before, that training an AI model on that content would introduce a bit of a bias We all know garbage in, garbage out. But liberal in, liberal out is an interesting idea, and I'm not sure how you fix it
It wasn't the training data, ChatGPT wasn't this bad at launch, it got worse as they "tuned it" to "reduce harmful content".
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#28I think most people would agree that a lot of the content on the internet is left-leaning. It seems obvious in hindsight, but I'd never considered it before, that training an AI model on that content would introduce a bit of a bias We all know garbage in, garbage out. But liberal in, liberal out is an interesting idea, and I'm not sure how you fix it
Always has been, from pot hole detection to crime prevention, &c. everywhere you use data you introduce bias and even worse, you have a good chance of perpetuating it: https://www.rand.org/content/dam/rand/pubs/research_reports/...
> I'm not sure how you fix it
I don't think you can, people are biased, people generated content is biased, these tools train on people generated content, there is no way to get an unbiased AI because unbiased opinions don't exists outside of pure maths/physics/&c. You'll never get an unbiased opinion about politics, or music, or culture
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#29Earlier quoted context omitted.
> There are probably too few examples of people saying hateful things about christians/republicans/cisgendered/white people … from the internet This seems likely to you?
Compared to the number of hateful things about Muslims/democrats/gays/blacks? Yes. I’d be actually shocked to find out the reverse. A good percentage of the US were born up in an era where you could legally bar people based on their race from your establishment. The extremely heated fight over gay marriage is still fresh in people’s memory. Trans rights are a controversial political issue where many mainstream politi…
Have you considered that your political bubble might minimize reporting of hate and violence originating from your in-group, while amplifying reporting of hate and violence originating from your out-group?
Re: The unequal treatment of demographic groups by ChatGPT/OpenAI content moderation
#30It reminds me of the day I typed in Google : 'why are women so patronizing?'. All the results were about men patronizing women. I just checked and it's still the case.
That seems like a better tactic than asking my wife the same question!