2023-04-06 21:43 UTC
Replies (40)
-
@ngaylinn@tech.lgbt 2023-04-06 21:58
@DrewKadel@social.coop Love this. We want so badly for ChatGPT to produce answers, opinions, and art, but all it can do is make plausible simulations of those things. As a species, we've never had to deal with that before.
-
@Patashu@dragon.style 2023-04-07 03:06
@DrewKadel@social.coop It's so important for everyone to understand this
-
@PCOWandre@jauntygoat.net 2023-04-07 03:13
@DrewKadel@social.coop No introspection and only waiting to generate the next line of a conversation. Sounds like quite a few humans!
-
@ageha@tomo.airen-no-jikken.icu 2023-04-07 16:49
@Moon@shitposter.club @PCOWandre@jauntygoat.net @thatguyoverthere@shitposter.club @DrewKadel@social.coop there really isn't a hard distinction between "reasoning" and "string language tokens together using a mathematical model that people find pleasing", though "reasoning" of the kind happens in solving a maths problem is a matter of shuffling higher/lossier/more-general abstractions. it's easier to play ma-jan or chess or whatever when you have hard, discrete gamepieces to move around a low-res board, rather than sliding around little piles of goop in a puddle. and following only certain game rules and not drawing on other things you happen to have read ("this sounds math-y") requires sharply-defined partitioning as well because our resources are so limited, humans often abstract prematurely, leading to the platonist error of projecting neat little models in our heads onto a complex world that doesn't well fit them. but these models have the opposite problem, so that even with enormous databases and computing resources they're too inefficient at model abstraction (i.e. compression) to model the world effectively (point again this conversation of woman specialises in producing simplified models with similar performance at greater efficiency https://traffic.libsyn.com/secure/computingup/computingup-ep265-11.mp3
-
@ageha@tomo.airen-no-jikken.icu 2023-04-07 17:09
@Moon@shitposter.club @PCOWandre@jauntygoat.net @thatguyoverthere@shitposter.club @DrewKadel@social.coop "systemic reasoning" just means "made from loosely-coupled modules when different modules are tightly coupled they bleed over and affect one another in all sorts of ways. section them off from one another and you get "the pristine rules of logic
-
@ageha@tomo.airen-no-jikken.icu 2023-04-07 17:26
@Moon@shitposter.club @PCOWandre@jauntygoat.net @thatguyoverthere@shitposter.club @DrewKadel@social.coop it can be that implementing something such kinda partitioned "self-reflection system" to self-edit is necessary step for effectively constructing and revising such abstraction, sure. questions on implementation details like that are where i think the next step has to be, not just "doing more of the same
-
@ignaloidas@not.acu.lt 2023-04-08 20:24
@Moon@shitposter.club @thatguyoverthere@shitposter.club @DrewKadel@social.coop @PCOWandre@jauntygoat.net tbh I don't expect trying to get reasoning out of LLMs to have any success. François Chollet has made a pretty damn reasonable reasoning challenge for computers in ~2019 and best solutions as of now can solve only about 30% of the tasks, and the algorithms for that as of now have zero neural networks. https://pgpbpadilla.github.io/chollet-arc-challenge
-
@roland@giersig.net 2023-04-09 14:15
@Moon@shitposter.club And how does ChatGPT differ in that kind of "mathematical" thinking from the way most people think about mathematics? 🤔 Yes, ChatGPT is built as a text predictor. That's very similar in how humans think and act. We also are association machines. We predict the future all the time. 🤷♂️ @PCOWandre@jauntygoat.net @thatguyoverthere@shitposter.club @DrewKadel@social.coop
-
@roland@giersig.net 2023-04-09 19:38
@thatguyoverthere@shitposter.club Well, most human sports heavily rely on our capabilities to predict the future. We have to mentally keep track of the spacial position of several objects and predict where those objects will be in the next few seconds. If you want to feel that prediction engine in our heads in action, just close your eyes while walking and notice that you can go on for several seconds without the urge to open your eyes. @PCOWandre@jauntygoat.net @Moon@shitposter.club @DrewKadel@social.coop
-
@Julie@social.coop 2023-04-07 03:25
@DrewKadel@social.coop Has she made it public? And if so, would you mind sharing the link? I know a lot of people who really need to see this.
-
@Odiseo79@mas.to 2023-04-07 03:35
@DrewKadel@social.coop Great description. The bot described itself in one of my conversations with it:
-
@Kazinator@mstdn.ca 2023-04-07 03:42
@DrewKadel@social.coop I've been a fairly prompt engineer: mostly on time to meetings and reasonably quick to respond to questions.
-
@pronoiac@mefi.social 2023-04-07 04:33
@DrewKadel@social.coop alt text: Something that seems fundamental to me about ChatGPT, which gets lost over and over again: When you enter text into it, you're asking "What would a response to this sound like?" If you put in a scientific question, and it comes back with a response citing a non-existent paper with a plausible title, using a real journal name and an author name who's written things related to your question, it's not being tricky or telling lies or doing anything at all surprising! This is what a response to that question would sound like! It did the thing! But people keep wanting the "say something that sounds like an answer" machine to be doing something else, and believing it *is* doing something else. It's good at generating things that sound like responses to being told it was wrong, so people think that it's engaging in introspection or looking up more information or something, but it's not, it's only, ever, saying something that sounds like the next bit of the conversation.
-
@kyozou@sfba.social 2023-04-07 04:42
@DrewKadel@social.coop She’s right. And that being the case, why on earth is anyone using it? If it’s not providing correct information, WTH is the point? I just saw a post that cites info from ChatGPT about companies doing business in TN. As if it’s an authoritative source…
-
@alakest@mstdn.social 2023-04-07 05:05
@DrewKadel@social.coop This seems squarely in "Chinese Room" territory. (https://en.m.wikipedia.org/wiki/Chinese_room) The main troubling questions if you accept this approach: "Where *does* understanding/meaning reside?" and "What *specifically* makes humans distinctive?"
-
@MostlyTato@mstdn.social 2023-04-07 06:18
@DrewKadel@social.coop I used to work with someone that was extremely good at telling people what they wanted to hear. She was also a feckwit. She got dismissed eventually. Reminds me of AI. It's just a flimflam merchant for t'intetnet.
-
@grtyvr@infosec.exchange 2023-04-07 07:22
@DrewKadel@social.coop you just have to ask it and it will tell you explicitly....
-
@btaroli@federate.social 2023-04-07 08:47
@DrewKadel@social.coop Right. Fancier Eliza that you use massive resources in cloud and pay for. Um, OK. Next. @jasongorman@mastodon.cloud
-
@jr@social.anoxinon.de 2023-04-07 09:22
@DrewKadel@social.coop what many people also do not realize is that it has no consciousness like a real brain. So it can't for example realize it's totally making up things... IMHO this problem is very self made though, as media tries to let people believe ChatGPT and companions are real "artificial intelligence" and not just very good neural networks...
-
@ook_simon@todon.nl 2023-04-07 09:34
@DrewKadel@social.coop Exactly, I completely agree. These chat bots are mostly gaslighting-as-a-Service or confirmation-bias-as-a-Service. The only intelligence they have is what we ascribe to them.
-
@habuma@fosstodon.org 2023-04-07 10:57
@DrewKadel@social.coop Another way of saying the same thing: Every so often, someone starts a social media trend when you type “I am” into the message app on your phone and then tap the first suggestion given. I just tried it and got “I am not sure if you are aware of this but I have a friend who is a nurse.” This is a perfectly reasonable completion for, although it may or may not be true. (And in my case, I don’t think it is true…maybe some distant friend is a nurse, but I don’t think so.)
-
@oligneisti@social.linux.pizza 2023-04-07 11:01
⬆ This is a great point: ⬆ ⬇ "It's good at generating things that sound like responses to being told it was wrong, so people think that it's engaging in introspection or looking up more information or something, but it's not, it's only, ever, saying something that sounds like the next bit of the conversation." @DrewKadel@social.coop
-
@deskmisfit@blorbo.social 2023-04-07 11:23
@DrewKadel@social.coop What's driving this misconception though is that most articles you find about these "AIs" are framing tools like ChatGPT as something that it is not. It collects data, rearranges it and spits it out. Whatever it gives you can sound perfectly sophisticated, but be wrong entirely, because it does not possess the intelligence most assume it should have, because of our human definition of "intelligence" and exorbitant use of the term "AI" without properly defining it.
-
@Npars01@mstdn.social 2023-04-07 11:33
@DrewKadel@social.coop Like Spotify's "sound-alike" song technology it's a product meant to sound like the original but isn't, ChatGPT is used as a cheat. Sounding like the original music artist is a cheat. It's not what people paid for. It isn't what people want. Fake music. Sounding like a human to attack elections for example, is an election cheat. It's not what people people paid for and it's not what people want. Fake democracy.
-
@URBANI@urbanists.social 2023-04-07 11:59
scary we're not ready @DrewKadel@social.coop
-
@pleaseclap@urbanists.social 2023-04-07 12:15
@DrewKadel@social.coop I think there's a difference between "being tricky" (tricking people by nature) and "playing tricks" (tricking people on purpose) eg. "deceptive" and "deceiving" are not the same thing - the former can exist without intent
-
@irisRichardson@mastodon.art 2023-04-07 12:32
@DrewKadel@social.coop Well stated. I tried ChatGPT after my son asked me to. I used my artist bio as my test text. I instructed it and it spits out a text with all I put in incorporated. For the most part, it kept my original bio text with some rearranging. It added some parts which applied and I had not thought of adding. But then it added things that did not apply to me at all. The user has to bring the knowledge to the table to know what is real and what is fiction.
-
@soeren_auer@mstdn.social 2023-04-07 12:43
@DrewKadel@social.coop my impression is it's the same with us humans. We also respond often with answers, which sometimes even don't even sound like right answers. I would say Llama function a lot like our brain.
-
@PixelJones@mindly.social 2023-04-07 12:43
@DrewKadel@social.coop Calling #ML, #ChatGPT machines that are designed to "say something that sounds like an answer" is the most accurate description I've heard. Unfortunately, to our peril, people are machines that are quick to conjure meaning from next to nothing (see tea leaves, bigfoot, Q-Anon), much less from well-formatted, grammatically correct streams of credible factoids confidently babbled out by witless word-stringers promoted as #AI.
-
@TransitBiker@urbanists.social 2023-04-07 13:42
@DrewKadel@social.coop it’s not “ai”
-
@TYB@mstdn.social 2023-04-07 13:45
@DrewKadel@social.coop So in essence, it's fox news ?
-
@morlando@econtwitter.net 2023-04-07 14:00
@DrewKadel@social.coop “Say-something-that-sounds-like-an-answer machine” Excellent. I'm using that from now on.
-
@log@mastodon.sdf.org 2023-04-07 14:21
@DrewKadel@social.coop In this, AI has replicated the least useful aspect of human experts. Quite a lot of them don't actually know what they're talking about either.
-
@zinelib@glammr.us 2023-04-07 14:43
@cristy@glammr.us
-
@aunteef@mstdn.party 2023-04-07 15:03
@DrewKadel@social.coop excellent explanation!
-
@MikeBon@mstdn.social 2023-04-07 15:10
@DrewKadel@social.coop but who needs something that just tells you what an answer to your query might sound like?
-
@inthehands@hachyderm.io 2023-04-07 15:17
@DrewKadel@social.coop @rjmccall@hachyderm.io This is particularly excellent because it both gives a fairly accurate non-technical heuristic for the general public to reason about the thing and at the same time is also a technically accurate summary of the actual mechanism. Hats off.
-
@Susan_Larson_TN@mastodon.online 2023-04-07 15:45
@DrewKadel@social.coop You have to verify information it gives you like any other source. Don't take what anyone tells you at face value.
-
@RupertReynolds@hachyderm.io 2023-04-07 16:01
@DrewKadel@social.coop Now that's what I call clear thinking :-)
-
@hadilq@androiddev.social 2023-04-07 16:26
@DrewKadel@social.coop that is totally correct in the same way that any conscious being would respond like that, trying to sound like the answer. Give it enough information about reality, not only abstract language, then it'll respond like us or better than us.