It’s because all the verbs are ill-defined in the context of AI. All it boils down to is “I don’t like LLMs”. You could say that LLMs can know, learn, and understand, and you’d be just as correct, just with opposite vibes.

They do learn by reading their corpus, and they do know what's in their corpus. If you created a temperature sensor and the AI removes itself from hot temps, then it feels pain. The insistency that AI not be antropormorphized sometimes gets in the way of communication.

> it feels pain

No... It detects temperature and reacts. It does not feel pain. Feels implies both affect and sensation, both of which are experiential, phenomenological qualia requiring some corollary of a nervous system and sentience, neither of which exist within the underlying structure of an LLM or indeed any current AI. AI can no more feel pain than a sliding door can decide to open. It's deterministically reacting to stimuli.

Dawkins faces this argument in "The selfish gene" and he choses to explicitly acknowledge that qualia is unprovable, so takes a definition of the phenomenon that is independent of it.

And this is when studying actual living organisms, so applying it to machines will be even more useful.

It's being downvoted by the AI's. They have feelings too you know...

Don't anthropomorphize AIs. They hate it when you do that.

It's being downvoted because it's the opposite of helpful and curious.

Go read this discussion I had with GPT-4:

https://gitlab.com/-/snippets/2594242

I wanted to get a feeling for the distinction philosophers used between "syntax" and "semantics", as well as exploring what someone in that camp would say about "concepts" and "understanding". The conversation seemed rational; just to make sure it wasn't way off base, I sent it to a friend of mine who had trained in philosophy, and he said he agreed with GPT-4 (and specifically, he didn't say "that's not what those terms mean in philosophy at all").

Now consider the sentence:

$SUBJECT $VERB the philosophical $NOUN of syntax and semantics well enough to apply it to my example of an adder circuit.

If $SUBJECT was "my nephew", then the most natural words to use for $VERB and $NOUN are "understood" and "concepts". What words should we use when $SUBJECT is "GPT-4"?

I submit to you that the most natural words to use in the context remain the same.

And in fact, "Of all places on the internet", I'd expect a forum filled with programmers to be very "functionalist". We spend our entire lives bridging the gap between "syntax" and "semantics" -- taking mechanical interactions and creating meaning out of them.

Unfortunately both of you are being downvoted because what the OP said is the truth about these stochastic parroting systems, and the AI bros here are upset and are fiercely deflecting the findings in this paper.

> Of all the places on the Internet, this is where I expected to see most people with minimal AI knowledge.

It should be. However, you're not far from a ChatGPT wrapper project that needs to get enough hype and attention to get the VC funding they need to execute their side grift.

Any 'technologist' that knows beyond the basics of AI knowledge will find that they don't trust these LLMs since they can't reason or understand their outputs and just spit out nonsense. This paper shows it is not even comparable to humans. It is not the same thing either.

Only a matter of time to search for an adversarial prompt to confuse these systems reliably.