5 ms·
> * As Pope Francis observes, “the very use of the word ‘intelligence’” in connection with AI “can prove misleading” Yes, LLMs are more about knowledge than in
by timeon 1y ago
> * As Pope Francis observes, “the very use of the word ‘intelligence’” in connection with AI “can prove misleading”
Yes, LLMs are more about knowledge than intelligence. AK rather than AI.
- diggan 1y agoIllustrating perfectly how wide this conversation really is, as we don't even have consensus about what "knowledge" means :)
- moralestapia 1y agoWe do! https://dictionary.cambridge.org/dictionary/english/knowledge https://dictionary.cambridge.org/dictionary/english/knowledg...
- diggan 1y agoThe Cambridge Dictionary has its own internal consensus, true, but there are so many more ways people understand that specific word :) Wikipedia even has it's own page with some of the various definitions people use: https://en.wikipedia.org/wiki/Definitions_of_knowledge https://en.wikipedia.org/wiki/Definitions_of_knowledge Then we have implicit/explicit knowledge (https://en.wikipedia.org/wiki/Implicit_and_explicit_knowledge https://en.wikipedia.org/wiki/Implicit_and_explicit_knowledg...) where some people assume one of them when they say "knowledge", others refer to the other. In fact, there is an entire scientific field to understanding what "knowledge" actually is/means: https://en.wikipedia.org/wiki/Epistemology https://en.wikipedia.org/wiki/Epistemology So yeah, it isn't as simple as looking it up in a dictionary, unfortunately.
- moralestapia 1y ago[flagged]
- fwip 1y agoYes, but blue doesn't have a "Definitions of Blue" Wikipedia page. There are nuances to definitions of common words "what is blue, what is a bicycle, what is a dollar, really?", but the magnitude of variance in definition is not shared with something like "knowledge" or "intelligence." With these high-level concepts, most people are operating only on a "I know it when I see it" test (to reference the Supreme Court case on obscenity).
- moralestapia 1y ago>Yes, but blue doesn't have a "Definitions of Blue" Wikipedia page. Oh, I understand, so the criteria is to have a Wikipedia page like that? You know what's interesting, I couldn't find neither of these: * تعريفات المعرفة * 知識嘅定義 * Définitions de la connaissance * Definiciones de conocimiento Should we add "and it has to be written in English" as a requirement? I know this is arguing ad absurdum, but the point is, again, that if you choose to be that strict, you wouldn't even be able to communicate with other people, because your desired perfect 1:1 map of concepts among them doesn't even exist.
- fwip 1y agoNo, I mean to illustrate that "blue" and "knowledge" have a vastly different degree in variation in definition. Like you say, all words of course have different definitions between individuals, but you and I are obviously able to communicate without specifying every definition. There exists a spectrum between well-agreed-upon definitions (like "and") and fuzzier ones. The definition of "knowledge" is divisive enough that many people disagree vehemently on definitions, which is illustrated by the fact that there is a whole Wikipedia article on it. If there is a "midwit trap" related to this, there is certainly a Sorites paradox trap as well - that because all words have varying definitions, that it is no use to point out that some words' definitions are more variable than others.
- karaterobot 1y ago> At some point you will understand that you will never have absolute and complete axioms from which to build everything on [1], and you have to work with what you have. To have hardware that displays blue, and code that manipulates blue, you must have a very clear and unambiguous definition of what blue means. Notice I did not say correct, only clear and unambiguous. Your whole point seems to be that words mean what a native speaker of the language understands them to mean, which is useful in linguistics and in the editing or dictionaries, but the context of this discussion is the representation of some concept in symbols that a computer can process, which is a different thing. Indeed, it's possible that the difference between code and 'vibes' will have to be in some way addressed by those very definitions of knowledge and intelligence, so I think these are relevant questions that can't be hand-waved away.
- throwup238 1y agoBecause everyone knows that millennia of epistemology can be summarized in a single incomplete sentence.
- abruzzi 1y agoI was always taught that knowledge was "justified true belief."
- glutamate 1y agoThe Gettier examples disagree!
- prometheus76 1y agoNow you get into the tricky waters of defining "justified" and "true". It's a circular definition that does not settle anything.
- ForOldHack 1y agoEinstien said it was what was left over when you forgot everything you leaned in school.
- slowmovintarget 1y agoKnowledge is well described (not necessarily explained, but described) in information theory. Intelligence, sentience, consciousness, even whether something is alive, are fuzzy concepts. Biology has a working definitions of "living organism" that includes a way to calculate likelihood that something is a living organism, but it still is probabilistic. Understanding is another concept that depends on philosophy of the mind as opposed to concrete physical processes.
- mock-possum 1y agoKnowledge, data, information, it’s all just awareness.
- shiandow 1y agoInteresting, I would consider knowledge to be something innately human in a way that solving problems isn't. Though intelligence is possibly even less well defined than knowledge, so it's hard to tell.
- lobsterthief 1y agoI believe knowledge is what you know based on facts and experience; wind sensors could gather data and store it in a database without a human touching it beyond initial setup. With enough data, and basic information about where the sensors are located, the computer becomes very knowledgeable about wind in a region without human intervention. I believe intelligence goes beyond that: knowing that such a system is a solution to an observed problem, architecting said system, using the output to solve a problem, analyzing the results, and deciding where to deploy additional systems. I think both examples above can be done by AI (if not now, then soon)—but only after being prompted carefully by a human. However, a generalized AI that can do all of the above for any problem in the known universe is likely very far off.
- cmrx64 1y agoif knowledge is a justified true belief, i’m down for saying LLMs have beliefs. to the extent that they are incorrigible, their faith may actually be superhuman.
- carabiner 1y agoNoam Chomsky just called it "plagiarism software."
- chpatrick 1y agoPretty disappointing from someone who spent his career modelling language and the cognition behind it.
- f30e3dfed1c9 1y agoIt is probably exactly because he spent a career considering the cognition behind language that he is not as impressed by LLMs as many others are. I'll readily admit to being expert in neither language and linguistics nor AI, but I am skeptical that anything going on inside an LLM is properly described as "cognition."
- drdaeman 1y agoI suspect the disappointment wasn't about whenever LLMs exhibit cognitive-like properties or not, but rather about the negative connotations tied to the word "plagiarism". Yea, they replicate patterns from their training data. So do we (ok, to be fair I have no idea about others but I believe I know that I do), and that's normal.
- permo-w 1y agodoes it really matter if it can be described as cognition or not? to me these models are useful for how effective they are, and that's literally it. the processes going on within them are extremely complex and at times very impressive, and whether some arbitrarily undefined word applies or not does not really matter. I think sometimes people forget that words are not maths or logic. when words come into language, no one sits down and makes sure that they're 100% logically and philosophically sound, they just start to be used, usually based on a feeling, and slowly gather and lose meaning over time. perhaps when dictionaries were first written there was some effort to do this, but for lots of words its probably impossible or incredibly difficult even now, never mind 200 years ago, if they could even be bothered in the first place. to give an example, a quite boring "philosophy question" that's bandied around, usually by children, is "if a tree falls in the forest and no one hears it, does it make a sound?". the answer is that "sound" is a word without a commonly-accepted, logically-derived meaning, for the reasons given above. so if to you the word sound is something human, then the answer is no, but if to you a sound is not something human, then the answer is yes. there's nothing particularly interesting or complex about the thought experiment, it's just a poorly defined word
- lupusreal 1y agoI think this is wrong. For nearly a hundred years, popular media has been priming the public to understand that artificial intelligence is superficially intelligent but is very prone to malfunction in inhuman ways. All that media in which AIs go haywire used to make nerds roll their eyes, but resonated with the general public then and has proved prescient now.
- lo_zamoyski 1y agoNot even knowledge. Knowledge requires intentionality and belief as well as justification. LLMs don’t have any of these. We ought to avoid anthropomorphizing LLMs. It is muddle headed.
- bmicraft 1y agoI'd add that knowledge usually implies a claim to truth, while llms can only offer information with varying likelihood of being true.
- lo_zamoyski 1y agoThis is what intentionality is about. No intentionality, no truth. An LLM doesn't deal with propositions, and it is propositions that are the subjects of truth claims. LLMs produce strings of characters that, when interpreted by a human reader, can look like propositions and result in propositions in the human mind, and it is those propositions that have intentionality and truth value. But what the LLMs produce are not the direct expression of propositional content, only the recombination of a large number of expressions of propositional content authored by many human authors. People are projecting subjective convention onto the objective. The objective truth about LLMs is far poorer in substance than the conventional readings we give to what LLMs generate. There is a good deal of superstition and magical thinking that surrounds LLMs.