this post was submitted on 25 May 2025

-8 points (30.0% liked)

Nowhere Else To Share

787 readers

13 users here now

I didn't know which community to post something in and so here we are.

Please comment if you know a more appropriate community for a post.

founded 2 years ago

MODERATORS

can@sh.itjust.works

can@lemmy.ca

ModeratorCan@lemmy.world

-8

Google AI is gender biased. (infosec.pub)

submitted 4 days ago by rainrain@sh.itjust.works to c/nowhereelsetoshare@sh.itjust.works

22 comments fedilink hide all child comments

all 23 comments

sorted by: hot top controversial new old

[–] Soliae@lemm.ee 13 points 4 days ago (1 children)

Tell me you don’t understand “AI” in LLMs without telling me you don’t understand AI in LLMs.

[–] tacosanonymous@lemm.ee 9 points 4 days ago (1 children)

Well, yeah. It’s an LLM built on a lot of scraping social media.

[+] rainrain@sh.itjust.works -7 points 4 days ago (1 children)

Now that's an interesting idea.

We ask questions. Get answers by scraping social media.. The answers inform social media. Informing further answers to questions. Etc. A spinning wheel.

It's deeply incestuous. In a hundred generations what monsters may spawn?

[–] irmoz@lemmy.world 3 points 3 days ago (1 children)

Its not "an interesting idea", it's exactly how they work.

[–] rainrain@sh.itjust.works 2 points 3 days ago (1 children)

Ideas about how things work can be interesting too.

[–] irmoz@lemmy.world 5 points 3 days ago* (last edited 3 days ago) (2 children)

Its not an "idea about how it works". It is how it works.

Can you not just admit you learned something here? Or do you just have to argue with everything to try and appear right?

What's wrong with, "oh, I didn't know that. How interesting!"

[–] can@sh.itjust.works 2 points 3 days ago (1 children)

I think it was the part about how the training data gets poisoned that was the interesting idea.

It is also the reality we're living in however.

[–] irmoz@lemmy.world 1 points 3 days ago* (last edited 3 days ago)

Based on their behaviour, I'm not so sure. It seemed to me to be a way of saying "that's maybe not true, but it's fun to think about". At least, that's how I'd use the phrase "that's an interesting idea". If I just found it interesting, I'd say "how interesting!"

But yes, it is indeed fascinating how LLMs work.

[–] rainrain@sh.itjust.works -2 points 3 days ago (1 children)

are you a robot?

[–] irmoz@lemmy.world 2 points 3 days ago (1 children)

Let's rewind to before that desperate (and likely spontaneous) accusation, and I'll give you another chance to reply in a normal manner.

No deflection. Just admit you didn't know LLMs scrape social media. That's all. It's okay; we don't come into this world with all of its knowledge.

[–] rainrain@sh.itjust.works -1 points 3 days ago (1 children)

I actually did have a vague idea in that general direction.

But that's rather beside my point. I mean, the AI definitely offered these answers. The answers are definitely gender biased. Offering that it's merely an artifact of the LLM technology is definitely a terrible excuse for that.

And given that LLMs are well known to be tweaked to align better with the philosophical styles of the hour, doubly so.

[–] irmoz@lemmy.world 3 points 3 days ago (1 children)

So, once again, you double down. No, you obviously didn't know that, and clearly still don't actually understand it, since you're claiming it was engineered to push a narrative (what narrative you won't say, but i bet it rhymes with bliss gandry).

Lastly, it's not an "excuse". It's an explanation. Calling it an "excuse" is just another attempt to deflect the answer and avoid being wrong.

[–] rainrain@sh.itjust.works -3 points 3 days ago (2 children)

Well given that the bias is generally considered a bad thing, explanations absolving them of responsibility for the badness are what is generally called an "excuse".

[–] irmoz@lemmy.world 3 points 3 days ago

It's not absolving them of responsibility. They're responsible for how they train their data - it doesn't need to be trained on social media.

[–] can@sh.itjust.works 5 points 3 days ago* (last edited 3 days ago) (2 children)

Unlocking because I don't want to stifle legitimate discussion

I don't know what to make of this post.

They inherit the biases of society (unfortunately).

They are also seeded with a bit of randomness for variety in responses. See this (YT Short)

Ask the same exact question and you'll get a variety of answers.

Edit: I'm beginning to think no more good discussion can come from this.

[–] irmoz@lemmy.world 1 points 3 days ago

Man I just don't get this guy.

[–] rainrain@sh.itjust.works -3 points 3 days ago (1 children)

Don't know what to make of this post?

Sounds more like you have very definite ideas about it. And are offering excuses on google's behalf.

[–] can@sh.itjust.works 7 points 3 days ago* (last edited 3 days ago)

Why would I, a Lemmy user, be defending google? Is that the most likely scenario here? Or is it that I just think all of these models are flawed in a similar way? The linked short was about ChatGPT.

Edit:

Maybe try posting this to !fuck_ai@lemmy.world ?

[–] moakley@lemmy.world 5 points 4 days ago

That's how people use those words. A woman is described as "pushy" for being assertive. It's a commonly known problem.