Geoffrey Hinton, despite his long history in the field, and having left Google r

Geoffrey Hinton, despite his long history in the field, and having left Google recently to caution against AIs, makes a rather weak (silly) set of arguments to justify his fear of AI.

Language models are quite knowledgeable but extraordinarily dumb at present, non recursive, non-predictive, unable to form world models, and unable to make moral decisions (reciprocity, property, demonstrated interests), and are being trained to get worse by humans trying to teach language models not to be offensive, and therefore to lie.

So instead of explaining how to move language models into world models, predictions, cause-effect models, and moral (legal) judgements, he’s just launching scare tactics because of what he sees.

Exasperating.
#GPT #AI #LLMs
https://t.co/dJDyJUTh1b