Location: About 4000 miles from the center of the earth (+/-100 mi)
Posts: 9712
Good Answers: 1113
Can Large Language Models (LLMs) Understand?
08/21/2024 12:01 PM
The Artificial Intelligence (AI) of the 1980's was carefully programmed with facts and reasoning logic. From explanations I have read and seen, the present day AI is different in that it just strings together words selected by the probability that one word will follow another, calculated from a huge amount of data copied from the internet.
It doesn't seem like something like this would work, yet AI seems to pass the "Turing test", and people seem to feel that it "understands". (I remember the Aliza program written in the 1960's that seemed intelligent at first, but could be tricked into spouting nonsense with a little effort.)
So, is today's AI just Eliza on steroids? Or on the other hand, is human intelligence just a regurgitation of events that we have sensed in our lifetime, similar to what AI is doing?
A lot of what we know, is based on observation, physical manipulation and comparison...I don't think AI has any cache of independently acquired knowledge that is not preprogrammed...The knowledge of what is true and what's false, is an unknown for most humans, let alone a computer chip, this inability to distinguish right from wrong is a fatal flaw...To be conscious you must be able to accurately analyze reality as you perceive it...this requires some measure of feedback from our senses...it must pass the smell test haha...When the computer has the responsibility to pay for its own electricity, then it might achieve some motivation for seeking truth, when your very existence depends on it, things become a lot more real...at least that's the way I see it...
__________________
All living things seek to control their own destiny....this is the purpose of life
This is the problem with AI, people want to make it conscious, for it to be conscious it must be alive, and once it's alive, it will assume all control over its own existence, and humans will become an afterthought...be careful what you wish for...
__________________
All living things seek to control their own destiny....this is the purpose of life
I've done a bit of reading on the subject, and the best description I've seen about how the brain works is that we build a working model (a simulation) of the outside world, including our selves, other people.
The brain receives signals from the senses and predicts what will come next. When information that contradicts the model comes in, the model is refined. I think that true AI would have to have this kind of design.
The existing AI doesn't learn (modify its model). It will never know more than what it has been trained on. Freezing the training is done on purpose. If it were allowed to learn after training and think for itself, the creator would lose control. (Anyone who's a parent understands this well.)
Here's a good book that I think is well worth reading.
That’s interesting… shook some cobwebs from my attic…
back in the late 1990’s, Fuzzy Logic which begin in the 1960’s came to the forefront, about applying fuzzy logic to AI because it’s logic system was simular to the how humans think… it got traction for a while and then disappeared… more then likely because processing speed was following Moores law.
__________________
“ When people get what they want, they are often surprised when they get what they deserve " - James Wood
This is purely conjecture, but I don't believe that large language models of AI have yet to understand communication. I also think that one day they will understand.
(Now if I had used an AI my grammar would have been better.)
__________________
"Don't disturb my circles." translation of Archimedes last words
Can Large Language Models (LLMs) Understand?
Re: Can Large Language Models (LLMs) Understand?
Comments rated to be "almost" Good Answers: