Are AI models doomed to always hallucinate?
techcrunch.com
techcrunch.com
(there so much math understood about this on so many level of analysis that it baffling as to why this is even asked).
The first contention, the better we get to approximate human intelligence the worse it will get.
The second contention, ultimately there will always be a political will to bend the truth or narrative (historical as well as current). And if they cannot do it at the source (ie. controlling the machines) they will simply legislate it, which is what has happened since COVID times, as well as with past, and/or recent armed conflicts. Other thoughts/facts/narratives are outlawed in so many degrees and categories of our collective life..., which might be acceptable on some level, what is nauseating however, is how inconsistently and how self-righteously they are applied in double standards.
The worst part is you don't even know why they hallucinate. Not even the 'AI researchers' can tell you why.
That is how opaque these systems are.
as long as people talk about language models as if they lie or make things up, this increases the weight of such an idea in any future corpus.
the models will get increasingly good at predicting “lies” because “they do.”
the panic will generate higher volume than level heads, skewing the output.
buckle up, boys and girls. humanity has no chill and the machines are heating up.