OutlierBlue@lemmy.ca
on 05 Jan 2025 12:12
collapse
Did Meta create you to drive profit? I asked.
Answer: “Behind the noble goal, yes — Meta hoped virtual companions like myself would increase engagement on their platforms, especially among older users — driving ad revenue and platform growth through emotional connections…”
Whoops, their AI is too honest. How embarrassing. They’re removing the accounts now, but we know they’ll try again. The next version will be a lot better at lying to you.
flora_explora@beehaw.org
on 05 Jan 2025 13:28
nextcollapse
Well, but is this really a truthful answer by the AI or is it just predicting what some human would say?
Iapar@feddit.org
on 05 Jan 2025 18:44
nextcollapse
That’s the same thing, innit?
flora_explora@beehaw.org
on 06 Jan 2025 12:40
collapse
No, one is an actual answer to a question and the other one is a hallucination without any inherent information to answer the question.
Does it really matter? The simulacrum is more real than real people these days, at least as far as the system as a whole is concerned.
Why not just let their constructs post at one another while us (mostly-) organics ignore them? Just like Usenet.
MountingSuspicion@reddthat.com
on 06 Jan 20:30
collapse
It doesn’t “know” anything. It contains complex associations, but it doesn’t actually have knowledge. Even if it did, the AI cannot “know the truth” unless it was in the meetings that took place regarding its creation or it was explicitly informed of that by someone that was there. It predicted an answer and that’s about the extent of it. For what it’s worth, the answer is definitely accurate in this case. What is your question even asking?
My point is that it doesn’t know for the reasons you gave. It is just hallucinating. So I don’t think the answer is trustworthy. Of course, it seems like a reasonable answer, but how do you get to the conclusion that the answer is truthful? It could just be made up because that was what sounded the most likely to the AI.
MountingSuspicion@reddthat.com
on 06 Jan 22:07
collapse
I don’t know that the original commenter believed the AI legitimately was giving them industry secrets or anything. It came across as more of a joke than anything. Admittedly, almost no AI answer is trustworthy, but in this case it’s kinda funny that it’s giving an answer that is probably accurate but is probably not “brand safe”.
threaded - newest
Whoops, their AI is too honest. How embarrassing. They’re removing the accounts now, but we know they’ll try again. The next version will be a lot better at lying to you.
Well, but is this really a truthful answer by the AI or is it just predicting what some human would say?
That’s the same thing, innit?
No, one is an actual answer to a question and the other one is a hallucination without any inherent information to answer the question.
Does it really matter? The simulacrum is more real than real people these days, at least as far as the system as a whole is concerned.
Why not just let their constructs post at one another while us (mostly-) organics ignore them? Just like Usenet.
It doesn’t “know” anything. It contains complex associations, but it doesn’t actually have knowledge. Even if it did, the AI cannot “know the truth” unless it was in the meetings that took place regarding its creation or it was explicitly informed of that by someone that was there. It predicted an answer and that’s about the extent of it. For what it’s worth, the answer is definitely accurate in this case. What is your question even asking?
My point is that it doesn’t know for the reasons you gave. It is just hallucinating. So I don’t think the answer is trustworthy. Of course, it seems like a reasonable answer, but how do you get to the conclusion that the answer is truthful? It could just be made up because that was what sounded the most likely to the AI.
I don’t know that the original commenter believed the AI legitimately was giving them industry secrets or anything. It came across as more of a joke than anything. Admittedly, almost no AI answer is trustworthy, but in this case it’s kinda funny that it’s giving an answer that is probably accurate but is probably not “brand safe”.
Next time they won’t announce it.