@Jobu Tupaki that's why someone is in quotes, I've gone from understanding people being frustrated with AI to outright irate at people trying to constantly claim victory over things it's not even designed to do...
It exhibits some cognitive functions, "intelligence" is just a junk term that most people use to mean either "magic" or "just plain human". It's not designed to do math, it doesn't have any of the cognitive abilities to actually process math. The fact that it can actually figure out the correct math functions half the time is hugely impressive, or the fact that it knows when asked to use python that floating point errors are a thing and will affect the operation because it contains a floating point value.
If you're going to complain about issues around artists, I'll leave you alone aside from minor factual statements. But the whole "AI is useless" or "AI is the new NFT" angles are just speaking out your ass. (To be clear, NFTs as we know them are a bastardization of a function in new and awful ways, the underlying function being little better than a thought experiment with little practical application)
For something that straight up has value on a variety of different fields, let alone the effect it has on providing disability accommodation tools (and that's not hypothetical, I've used them for my own accommodations), I'm simple at the point of telling people to go fuck themselves for being high and mighty about it.
@shiri @RubyTuesdayDONO On that basis, LLMs can't be criticized for their output at all, because the only thing they are "designed to do" is create answers that look like they could have been written by a human.
"Oh, you got severely poisened by chlorine gas while trying to clean your washing machine? Well, shouldn't have done what LLM suggested, it's not designed to give advice for cleaning washing machines!" 👏
We know LLMs are ill-suited to doing maths. The problem is that a good chunk of people don't know so — either believing they can replace their entire workforce with it or that it will lead us to AGI, hence why we are shouting from the rooftops about it. People having tempered expectations about what LLMs can do is fundamentally a good thing.
@shiri @purple But nobody expects notepad to play video. That *is* the entire point here. It's true that people generally don't know what LLMs are, and the ridiculous marketing of them doesn't help, so these examples are great at pointing out their limitations. We need more of this, not less.
@shiri @veronica @purple @jmopp What makes this evidence of generative AI being garbage is that it can't do maths but confidently gives you incorrect answers anyway, and even doubles down when challenged.
Notepad can't play videos, no, but if you try to make it load a video file anyway it quickly and clearly becomes apparent to the user that it can't play videos.
Generative AI also has a limited scope of things that it can do (e.g. SEO filler content & misinfo bots) but unlike Notepad it will attempt to do everything else too, and confidently serve up nonsense with no easy way for the user to know when this is happening. That's a fundamental difference and make it deserving of the label garbage, IMO.
@veronica @shiri …and nobody has ever marketed notepad as the ‘solution’ to video, or really anything for that matter, whereas major companies are spending millions telling us LLMs are the answer to pretty much everything, including making videos and video players, but also government and healthcare…
@purple @shiri It's stupid to say that "AI is garbage" and post an image of an LLM getting math wrong. It is functioning pretty well as far as I can tell.
@shiri Large Language Models are not intelligent, their perceived intelligence is part of the language itself, not of the models. #LLMs don't understand anything, they just memorise patterns and generate synthetic patterns that look similar to the ones they have learned. There is no mind in there, no reasoning.
When defining intelligence you're either going to exclude a lot of people you would consider intelligent, or you're going to have to accept that we're not the only ones with intelligence.
I'm willing to accept that we're not the only ones, and the definition I accept is "possesses cognitive abilities", with my prime example being the cognitive ability that it is better at than me: Executive Function.
@Aaron Toponce ⚛️:debian: @Lord Caramac the Clueless, KSC Just for reference: I'm ADHD and that means my executive function is severely impaired, someone without impairment is probably better at it than the LLM, but in this case it's exhibiting a cognitive function that exceeds the capacities of some human beings.
It's one of the popular assistive uses of LLMs, see https://goblin.tools/ for that purpose.
@shiri I know, I myself am struggling with the very same thing, as you could easily tell if you saw me now sitting next to a huge pile of dirty laundry that should have been washed two weeks ago and a whole lot of unfinished projects, some of which haven't been worked on in years.
"When defining intelligence you're either going to exclude a lot of people you would consider intelligent, or you're going to have to accept that we're not the only ones with intelligence."
While what you say is true, it does in no way follow that LLMs can not not be intelligent.
LLMs are not intelligent, because they by design predict the next token in a deterministic manner.
It doesn't meet *ANY* sensible definition of intelligence, not even your own.
@shiri Chatbots aren't people, but when people go around confidently spouting stuff that's this wrong they are fair game to be mocked. An actual person with a disability who isn't able to answer these questions would just say "I don't know"
@shiri McDonald's fired their drive through AI because it kept screwing up orders. It's about as useless as a fart in a blizzard. There's nothing to defend.
@shiri it mimics a handful of cognitive functions? Which is very useful in some cases. It becomes less useful as it becomes less trustworthy in its output, but if the stakes are low you can probably rely on it to make life easier in lots of little ways.
@Xe :verified: @Aaron Toponce ⚛️:debian: yeah, but as far as I know it doesn't really have any capacity for computation, just association and linguistic understanding. And it sounded like you had insight on how it reached the conclusion of 0.21
@cadey In other words, despite all efforts to make math work better with LLMs, like adding Python support, it's still bad at it. Also it inherited the overconfidence from the dataset, which should include Reddit.
@Hexa there's always one promptfondler in the thread that doesn't understand that you can't get fully repeatable answers from the confabulation engine, and that any answer to that question is a valid answer within the llm paradigm, no matter if it's incorrect or not.
(there's also another promptfondler who thinks that the problem is just in one particular llm, not in the way llm works)
I wonder: you know how virtual assistants are given feminine names and voices (Siri, Alexa)? And you know how there is a persistant false belief that women are somehow worse at math than men?
I have to wonder whether that combination of biases has any influence on the programmers who create these LLMs? I mean on top of all of the other biases and misunderstandings they already have about neuroscience and language? Are they creating their own stereotype of a ditzy secretary?
For the example - ChatGPT botching arithmetic - it actually passed the Turing test. Once in a store, I ordered 2.2 lb of some deli item, and the scale registered 2.02. The guy behind the counter called 2.20 "two point twenty" and 2.02 "two point two". The scale always showed two digits past the decimal point. This guy basically made the same mistake as ChatGPT.
Shiri Bailem
in reply to Aaron Toponce ⚛️:debian: • •@Aaron Toponce ⚛️:debian: ... so you prove "someone" lacks all intelligence by challenging their math disability?
It only has a handful of cognitive functions, none of which are good at math. Honestly, it does better than I expected.
like this
aberl🏳️🌈✅, br00t4c, Kate (i ❤ blahaj) :unverified:, Acvaristul Lenes and Old Man in the Shoe like this.
Jobu Tupaki
in reply to Shiri Bailem • • •Shiri Bailem
in reply to Jobu Tupaki • •@Jobu Tupaki that's why someone is in quotes, I've gone from understanding people being frustrated with AI to outright irate at people trying to constantly claim victory over things it's not even designed to do...
It exhibits some cognitive functions, "intelligence" is just a junk term that most people use to mean either "magic" or "just plain human". It's not designed to do math, it doesn't have any of the cognitive abilities to actually process math. The fact that it can actually figure out the correct math functions half the time is hugely impressive, or the fact that it knows when asked to use python that floating point errors are a thing and will affect the operation because it contains a floating point value.
If you're going to complain about issues around artists, I'll leave you alone aside from minor factual statements. But the whole "AI is useless" or "AI is the new NFT" angles are just speaking out your ass. (To be clear, NFTs as we know them are a bastardization of a function in new and awful ways, the underlying function being little better than a thought experiment with little practical application)
For something that straight up has value on a variety of different fields, let alone the effect it has on providing disability accommodation tools (and that's not hypothetical, I've used them for my own accommodations), I'm simple at the point of telling people to go fuck themselves for being high and mighty about it.
like this
GhostOnTheHalfShell, DearFox, Jackie (aka Queen Antifa) 🌹:debian_logo::linux:, Max, Acvaristul Lenes and Old Man in the Shoe like this.
Max reshared this.
words_number
in reply to Shiri Bailem • • •@shiri @RubyTuesdayDONO On that basis, LLMs can't be criticized for their output at all, because the only thing they are "designed to do" is create answers that look like they could have been written by a human.
"Oh, you got severely poisened by chlorine gas while trying to clean your washing machine? Well, shouldn't have done what LLM suggested, it's not designed to give advice for cleaning washing machines!" 👏
Beyuum
in reply to words_number • • •words_number
in reply to Beyuum • • •purple is tired
in reply to Shiri Bailem • • •Shiri Bailem
in reply to purple is tired • •like this
Jackie (aka Queen Antifa) 🌹:debian_logo::linux:, Kevin Stewart, Acvaristul Lenes, kb and Simon Gray like this.
Shiri Bailem
in reply to Shiri Bailem • •like this
Jackie (aka Queen Antifa) 🌹:debian_logo::linux:, Demiurg, Kevin Stewart, hex and Simon Gray like this.
·J Mopp
in reply to Shiri Bailem • • •Veronica Olsen 🏳️🌈🇳🇴🌻
in reply to Shiri Bailem • • •Shiri Bailem
in reply to Veronica Olsen 🏳️🌈🇳🇴🌻 • •@Veronica Olsen 🏳️🌈🇳🇴🌻 @purple is surprised by the French @Aaron Toponce ⚛️:debian: this @·J Mopp I wouldn't bat an eye if it wasn't under the heading of "AI is garbage", OP is trying to argue that it has no value at all because it can't do this one thing.
If it was "Don't trust what AI tells you" or "AI isn't good at everything" I'd probably be boosting it instead.
hex likes this.
Anthony Horton
in reply to Shiri Bailem • • •@shiri @veronica @purple @jmopp What makes this evidence of generative AI being garbage is that it can't do maths but confidently gives you incorrect answers anyway, and even doubles down when challenged.
Notepad can't play videos, no, but if you try to make it load a video file anyway it quickly and clearly becomes apparent to the user that it can't play videos.
Generative AI also has a limited scope of things that it can do (e.g. SEO filler content & misinfo bots) but unlike Notepad it will attempt to do everything else too, and confidently serve up nonsense with no easy way for the user to know when this is happening. That's a fundamental difference and make it deserving of the label garbage, IMO.
Acvaristul Lenes
in reply to Anthony Horton • • •Prof Kemi FG
in reply to Shiri Bailem • • •@shiri @veronica @purple @jmopp
"Don't trust what AI tells you" is pretty darn close to "AI is garbage" tbqh
purple is tired
in reply to Veronica Olsen 🏳️🌈🇳🇴🌻 • • •OpticalNail 🇵🇸
in reply to purple is tired • • •Lord Caramac the Clueless, KSC
in reply to Shiri Bailem • • •Lord Caramac the Clueless, KSC
in reply to Lord Caramac the Clueless, KSC • • •Shiri Bailem
in reply to Lord Caramac the Clueless, KSC • •@Lord Caramac the Clueless, KSC @Aaron Toponce ⚛️:debian: age old argument that boils down to not understanding that intelligence is far fuzzier a concept than you're willing to accept.
When defining intelligence you're either going to exclude a lot of people you would consider intelligent, or you're going to have to accept that we're not the only ones with intelligence.
I'm willing to accept that we're not the only ones, and the definition I accept is "possesses cognitive abilities", with my prime example being the cognitive ability that it is better at than me: Executive Function.
like this
Demiurg, Max and Powersource like this.
reshared this
Max and dibi58 reshared this.
Shiri Bailem
in reply to Shiri Bailem • •@Aaron Toponce ⚛️:debian: @Lord Caramac the Clueless, KSC Just for reference: I'm ADHD and that means my executive function is severely impaired, someone without impairment is probably better at it than the LLM, but in this case it's exhibiting a cognitive function that exceeds the capacities of some human beings.
It's one of the popular assistive uses of LLMs, see https://goblin.tools/ for that purpose.
like this
Powersource and Joe Vinegar like this.
reshared this
dibi58 and Joe Vinegar reshared this.
Lord Caramac the Clueless, KSC
in reply to Shiri Bailem • • •Powersource
in reply to Shiri Bailem • • •Magic ToDo - GoblinTools
goblin.toolsFabian Transchel
in reply to Shiri Bailem • • •@shiri @LordCaramac
"When defining intelligence you're either going to exclude a lot of people you would consider intelligent, or you're going to have to accept that we're not the only ones with intelligence."
While what you say is true, it does in no way follow that LLMs can not not be intelligent.
LLMs are not intelligent, because they by design predict the next token in a deterministic manner.
It doesn't meet *ANY* sensible definition of intelligence, not even your own.
br00t4c
in reply to Lord Caramac the Clueless, KSC • • •Sofia ☭🇧🇷☭
in reply to Shiri Bailem • • •Sammy 🐾
in reply to Shiri Bailem • • •kami_kadse
in reply to Shiri Bailem • • •Steven Capobianco
in reply to Shiri Bailem • • •I don’t agree with this mindset.
If you sell a product designed for specific functions (computation in this case) and it does not do that specific function, then it needs more work.
Especially math where it’s very rigid with how problems are solved.
TrackerRoo :verified:
in reply to Shiri Bailem • • •missed_sla
in reply to Shiri Bailem • • •cholling
in reply to Shiri Bailem • • •Iwillyeah
in reply to Shiri Bailem • • •Abe Froman
in reply to Shiri Bailem • • •ferret
in reply to Shiri Bailem • • •Xe :verified:
in reply to Aaron Toponce ⚛️:debian: • • •Shiri Bailem
in reply to Xe :verified: • •like this
dibi58 likes this.
dibi58 reshared this.
Xe :verified:
in reply to Shiri Bailem • • •Shiri Bailem
in reply to Xe :verified: • •Gustavo
in reply to Xe :verified: • • •Ryan Dormanesh
in reply to Aaron Toponce ⚛️:debian: • • •ChatGPT
chatgpt.comflere-imsaho
in reply to Ryan Dormanesh • • •@Hexa there's always one promptfondler in the thread that doesn't understand that you can't get fully repeatable answers from the confabulation engine, and that any answer to that question is a valid answer within the llm paradigm, no matter if it's incorrect or not.
(there's also another promptfondler who thinks that the problem is just in one particular llm, not in the way llm works)
@atoponce
Grrrr, Darth Moose Shark reshared this.
Ryan Dormanesh
in reply to flere-imsaho • • •Emma (has_many_books of old)
in reply to flere-imsaho • • •Grrrr, Darth Moose Shark reshared this.
Noah Cook
in reply to Aaron Toponce ⚛️:debian: • • •I wonder: you know how virtual assistants are given feminine names and voices (Siri, Alexa)? And you know how there is a persistant false belief that women are somehow worse at math than men?
I have to wonder whether that combination of biases has any influence on the programmers who create these LLMs? I mean on top of all of the other biases and misunderstandings they already have about neuroscience and language? Are they creating their own stereotype of a ditzy secretary?
flere-imsaho
in reply to Noah Cook • • •Grrrr, Darth Moose Shark reshared this.
Bill Zaumen
in reply to Aaron Toponce ⚛️:debian: • • •flere-imsaho
in reply to Bill Zaumen • • •Erik Jonker
in reply to Aaron Toponce ⚛️:debian: • • •flere-imsaho
in reply to Erik Jonker • • •Erik Jonker
in reply to flere-imsaho • • •