This is totally on point. I told gemini explicit instructions not to make shit up, things it told me to say, and it totally hallucinated over and over. I asked if paid versions are better. Then it gave sources that yes, paid is better, but not foolproof. Basically, the paid versions sandbox the shit out of the answer and then have it verified by other “agents.” The problem with this is, once it has access to the internet, it’s fucked. Reddit has poisoned most of the internet. There is no way for it to consistently get the right answer. It saying it’s completely correct is wrong. The only good thing it’s for is a search engine to get random, cool sources. I’m sure they’re ruining that as we speak. Why can’t these greedy mother fuckers do what’s right for the world.
You want ai to do something it’s not able to do. Hallucinations are the foundation of how it generates text. Everything out generates is a hallucination, just some are more accurate than others.
This meme is not on point, it’s a crude simplification of something we have no idea about.
A super intelligence might not think about us at all, it might not even think it the same way we think.
On the other side, it might value everything humans have achieved including be the maker.
Just saying it’s going to be evil and set on destroying all humanity is superficial and boring.
You want ai to do something it’s not able to do. Hallucinations are the foundation of how it generates text. Everything out generates is a hallucination, just some are more accurate than others.
I agree with you, but that can be a useful tool in certain situations. They completely lied about its capabilities. A super intelligence will hopefully have a human and robot survival mode built in.
pelespirit@sh.itjust.works 1 week ago
This is totally on point. I told gemini explicit instructions not to make shit up, things it told me to say, and it totally hallucinated over and over. I asked if paid versions are better. Then it gave sources that yes, paid is better, but not foolproof. Basically, the paid versions sandbox the shit out of the answer and then have it verified by other “agents.” The problem with this is, once it has access to the internet, it’s fucked. Reddit has poisoned most of the internet. There is no way for it to consistently get the right answer. It saying it’s completely correct is wrong. The only good thing it’s for is a search engine to get random, cool sources. I’m sure they’re ruining that as we speak. Why can’t these greedy mother fuckers do what’s right for the world.
PixelatedSaturn@lemmy.world 1 week ago
You want ai to do something it’s not able to do. Hallucinations are the foundation of how it generates text. Everything out generates is a hallucination, just some are more accurate than others.
This meme is not on point, it’s a crude simplification of something we have no idea about. A super intelligence might not think about us at all, it might not even think it the same way we think. On the other side, it might value everything humans have achieved including be the maker.
Just saying it’s going to be evil and set on destroying all humanity is superficial and boring.
pelespirit@sh.itjust.works 1 week ago
I agree with you, but that can be a useful tool in certain situations. They completely lied about its capabilities. A super intelligence will hopefully have a human and robot survival mode built in.