Google's AI is still broken

I like the fact they have us saying "hallucination rates" when we talk about error and false information rates. We do like jargon.

A lot of people will accept this tech as authority, and while it may get better as it steals more human input, it will always be open to the manipulation of faithful followers.
 
img_1_1761863458435.jpg
 
That is one of the biggest problems with AI. In the above cartoon the AI answer is actually true as the shroom is actually edible in the sense that it can, in fact, be eaten even though it should not.
The cartoon is dead on. It's interesting that while AI has issues with the data it's fed, it also is a very literal system. That's very useful for working through enormous sets of stats, or medical/mathematical functions. It's less useful for interacting with non literal thinking humans - the majority of us.
When you follow the money it leads to AI, so we are going to have to adjust to it, or drop out of the online world. It most likely won't be that good at adjusting to us, except to manipulate us.
 
Edible doesn't mean "can be chewed up and swallowed". Edible means fit to be eaten. So the AI in the cartoon was incorrect.
 
I just had an amusing read. I am a sucker for any article about my hockey team - ice hockey, that is. I just read a piece that started out like it was talking to simpletons, with breathy, smooth, vacant prose. As it went on, it became very confused about the sport in question, also talking about field hockey. It then started using pronouns to reference people it had never introduced. "He said" without ever mentioning who he might be.

I notice sports writing by AI is having serious gender pronoun issues, and is interpreting many eastern European male names as female. This is the golden age. It'll get boring as it is improved.
 

Most reactions

Back
Top