In this study they asked to replicate 1:1 headline publisher and date. So for example if AI rephrased headline as something synonymous it would be considered at least partially incorrect. Summarization doesn’t require accurate citation, so it needs a separate study.
Comment on I totally missed the point when PeerTube got so good
Stillwater@sh.itjust.works 2 days agoIt might be wrong more often than you think
hisao@ani.social 2 days ago
Stillwater@sh.itjust.works 2 days ago
OK but google (or ask your AI?) about AI scurvy. This isn’t the only source saying theres a problem with the answers.
LesserAbe@lemmy.world 2 days ago
Besides the other commenter highlighting the specific nature of the linked study, I will say I’m generally doing technical queries where if the answer is wrong, it’s apparent because the AI suggestion doesn’t work. Think “how do I change this setting” or “what’s wrong with the syntax in this line of code”. If I try the AI’s advice and it doesn’t work, then I ask again or try something else.
I would be more concerned about subjects where I don’t have any domain knowledge whatsoever, and not working on a specific application of knowledge, because then it could be a long while before I realize the response was wrong.
thedruid@lemmy.world 2 days ago
IS wrong
Ftfy