Post

would love to hear even a single actual argument about how AI "changes the nature of humanity" not based in fanciful speculation

Demonstrating reasoning and language capabilities in computers alters our understanding of self, even if we don't particularly like the alteration

I disagree with your fundamental premise. The fact that a machine trained on billions of real conversations online can mimic reasoning does nothing to alter my understanding of self. And you haven't given any argument as to why it should.

This is the crux of the debate, right? Whether it's doing mimicry or reasoning. I struggle to see how you could view it as purely mimicry. It has to tie together multiple novel thoughts in a logical construction to create its responses.

when you speak to other people, do you gauge your words based upon the statistical probability that the next word you say is the word the other wants to hear, or do you gauge them based upon your own desires, interests, and opinions?

what do you want me to say here

re: mimicry vs reasoning. the machine chooses words based upon statistical probabilities that a word, phrase, or concept may link to a previous set of words. we don't do that when we communicate. we don't have statistics engines in our heads.

I'm arguing that this is not a machine capable of human reason because it does not form language the way that we do. it forms a facsimile of language through mathematical assumptions, whereas for us, language is the tool we use to understand and negotiate reality.

an LLM may look like it reasons, but it can't because it doesn't exist as a being structured in reality by language. it can only recreate the appearance of reason which is so far from the same thing as to be dangerously unreliable.

not to be uhh. controversial. have either of you learned how any of this works? (its completely fine if not) But like, linear regression, non linear solutions w/ conv networks, I’m talking the basis of neural networks that lead to modern LLM’s and why its statistically based and grounded in NLM

I boardly understand the ideas of the stacking MLPs, the attention heads, back propagation to train the weights, tokenization, vector embeddings, etc. But I am far from an ML or LLM researcher