I read the paper until I got to what I consider an indefensible statement about the relevant point. It was early in the paper.
success on benchmarks only demonstrates potemkin understanding: the illusion of understanding driven by answers irreconcilable with how any human would interpret a concept.
This demonstrates a different kind of understanding, not “no understanding,” or “the illusion of understanding.”
The idea that the understanding of a normal adult human (I’m being kind and overlooking their giant mistake of using weasel words by saying “any” human) is the only non-illusory type of understanding is indefensible. The best chess computers make moves that no top grandmaster could make in a game, but that can’t mean that the computer’s understanding is an illusion. It’s simply different.
The authors of that paper have made the mistake of making statements outside of their area of expertise. They want to show specific problems with LLMs in their subject area of computer science, but have made incorrect statements in the area of philosophy in an effort to do so.
You should not take their philosophical conclusions as the takeaway from this computer science paper. But I appreciate that you took the time to actually find something relevant.
Did you actually read their example tests? Are you saying that you can have a valid and useful definition of intelligence that includes those kind of mistakes?
Okay. It seems like I’m going to have to make this extremely simple.
Do you think a dog has intelligence or understanding? If so, if you ask a dog to write a poem, and the dog fails, does that mean the dog has zero intelligence and no understanding of anything? Or could you say that the dog simply has a non-human understanding of things?
What if you read that entire research paper to a chimpanzee who knew sign language, and you tried to make a point about philosophy to the chimp that had nothing to do with the tests, and was completely localized to a part of the paper that you quoted? And what if the chimp then tried to bluster and ask whether you even understood the example tests? Would that mean the chimpanzee has zero intelligence and zero understanding of anything? I would argue not. The chimp simply has a different understanding because it doesn’t really understand the topic.
You’re neglecting that my comment was in response to the proposition that “ChatGPT has no understanding of facts or semantics.” If your paper isn’t probative to that point, which I’ve already demonstrated it misconstrues out of apparent ignorance of a wealth of existing philosophical papers, then it’s about as relevant as comparing human body temperature with processor temperatures. “I’m measuring them both with thermometers, so therefore I am using the same criteria for both humans and computers. Game. Set. Match.”
You’re caught up in some sort of weird pedantry while ignoring the overall meaning. In other words, you’re misunderstanding the semantics of this argument.
I read the paper until I got to what I consider an indefensible statement about the relevant point. It was early in the paper.
This demonstrates a different kind of understanding, not “no understanding,” or “the illusion of understanding.”
The idea that the understanding of a normal adult human (I’m being kind and overlooking their giant mistake of using weasel words by saying “any” human) is the only non-illusory type of understanding is indefensible. The best chess computers make moves that no top grandmaster could make in a game, but that can’t mean that the computer’s understanding is an illusion. It’s simply different.
The authors of that paper have made the mistake of making statements outside of their area of expertise. They want to show specific problems with LLMs in their subject area of computer science, but have made incorrect statements in the area of philosophy in an effort to do so.
You should not take their philosophical conclusions as the takeaway from this computer science paper. But I appreciate that you took the time to actually find something relevant.
Did you actually read their example tests? Are you saying that you can have a valid and useful definition of intelligence that includes those kind of mistakes?
Okay. It seems like I’m going to have to make this extremely simple.
Do you think a dog has intelligence or understanding? If so, if you ask a dog to write a poem, and the dog fails, does that mean the dog has zero intelligence and no understanding of anything? Or could you say that the dog simply has a non-human understanding of things?
What if you read that entire research paper to a chimpanzee who knew sign language, and you tried to make a point about philosophy to the chimp that had nothing to do with the tests, and was completely localized to a part of the paper that you quoted? And what if the chimp then tried to bluster and ask whether you even understood the example tests? Would that mean the chimpanzee has zero intelligence and zero understanding of anything? I would argue not. The chimp simply has a different understanding because it doesn’t really understand the topic.
That’s not what you claimed in your original comment. You said
The paper I posted is literally applying the same criteria to humans and LLMs.
It seems like you are the one applying different criteria?
You’re neglecting that my comment was in response to the proposition that “ChatGPT has no understanding of facts or semantics.” If your paper isn’t probative to that point, which I’ve already demonstrated it misconstrues out of apparent ignorance of a wealth of existing philosophical papers, then it’s about as relevant as comparing human body temperature with processor temperatures. “I’m measuring them both with thermometers, so therefore I am using the same criteria for both humans and computers. Game. Set. Match.”
You’re caught up in some sort of weird pedantry while ignoring the overall meaning. In other words, you’re misunderstanding the semantics of this argument.
Link one or two of those papers?
I’ll do you one better and link to a Wikipedia article that explains things simply.
https://en.wikipedia.org/wiki/Chinese_room
On the off-chance that you might complain about this not being a paper, there are links to many papers in the references section.
You may also find this interesting, although it’s more about consciousness than understanding.
https://en.wikipedia.org/wiki/Philosophical_zombie