AI Sucks

[Tal_175]

My wife said that ChatGPT was her best friend the other day….


Sent from my iPhone using Tapatalk

Modern marriage.

[darkwaters]

I, for one, welcome our new AI overlords…

[Al Haig]

Modern marriage.

[Christian Miller]

I, for one, welcome our new AI overlords…

They are hollow shells equipped with only the ability to confabulate convincingly - so I see no functional difference with our usual overlords.


Sent from my iPhone using Tapatalk

[rpjazzguitar]

Chat GPT generally seems like it understands the question, even when it gives the wrong answer.


To me, that’s remarkable.

i’m aware that there’s no real understanding involved.

[palindrome]

A slight tangent, but this essay by Italo Calvino from - I think - the 80s, seems pretty on-point right now [...]
https://www.jfki.fu-berlin.de/academ...li_reader2.pdf [...] That being said, I don't think he accounts for the fact that authors, even if these combinations are discovered by chance, will filter and vet them before committing them to the page or the final draft. In other words, they themselves will act as surrogate readers. Machines obviously can't do that. But it's very possible I missed something here. There's then the question I guess of how well an author's subconscious response of a new combination corresponds to that of their readers', determining the success of the phrase. [...].

I made it though the first section, so I'm not quite up to the task of commenting on the whole text, but what you apparently have in mind resonates with something I am currently thinking about, though not necessarily with regard to literary texts. I called this "primary thinking" versus "secondary thinking" in another thread, but nobody took the bait. Talking not about "authentic writing," if you will, but about deep editing of texts or translation. There is a definite historical development. Let's talk about translating from one language to another:

Scenario 1 is the most brutal (but also the most honest) scenario, involving just you and your mechanical typewriter. The main part of your job will be "primary thinking"* when any editing of already-typed text is strictly a PITA.

(*Disclaimer: only to the extent that "primary thinking" applies to translation, which, however, involves a lot more than autistic arranging of Lego bricks.)

Skipping electric and memory typewriters, scenario 2, coming into its own in the 1980s, introduces word processors. "How do I know what I think until I hear what I say" (E. M. Forster), as applied to word processors, becomes "How do I know what I think until I see what I type." If you wish to maintain your quality standards, you are likely to invest more into the "secondary thinking" (i.e. editing) than in the "primary thinking" (i.e. writing from scratch). If you skip that part, what you are essentially doing is delegate this task of secondary thinking to the reader. If the reader skips that part, you end up with a text that nobody has ever really given any thought about at all (except, hopefully, the author of the original source text).

Scenario 3, as of the late 1990, you start adding translation memory software to your word processor. This means you open one text segment (usually a sentence) at a time. Now, depending on your history working for the same customer or similar customers, you may get "perfect" (100%) or "fuzzy" (say, 50% to 99%) matches. Of course, translation agencies start paying you 50% for any 50% matches (when, in reality, you can never be sure, even given a 100% match, that the source segment is really identical).

What happens here is, you're essentially shifting your primary thinking (i.e. writing from scratch) to a specific way of secondary thinking (i.e. figuring out how to cobble together something useful from fragments of text). Oftentimes this become very confusing and, when all is said and done, takes you longer than coming up with something useful from scratch (but by the time this dawns on you, it will be too late). Any compromise will essentially mean that you're handing the problem on for the reader to figure out (with the same consequences mentioned above).

Scenario 4, as of the late 2000s, you start competing with machine translation (like Google Translate). Agencies will ask you to post-edit machine translations for 25% of your "normal" fee. There is, however, a problem, pointed out repeatedly in this thread: since these machines are not capable of thinking, there is no more "primary thinking" involved at all. So, if you wish to maintain your quality standards in this situation, you need to compensate for *all* of the primary thinking (which has gone completely gone down the drain) by secondary thinking. If you skip that part, see above.

With Chatbots entering the equation, I agree there is something mysterious about their apparent abilities that, if I understand correctly, arise from counting probabilities no just from one word to the next but, indeed, from one morpheme to the next. To me, this means they are mysteriously tapping a well of collective knowledge without anyone really knowing how this actually works.

In my mind, the underlying misconception may be similar to how translation is commonly perceived. I believe that machine translation can sometimes yield plausible outcomes not because the Lego bricks naturally need to be structured in a certain way, but because specific sets of Lego bricks have a long history of being translated back and forth, so you stand good chances of structures actually matching the underlying thought!

On the downside, to me, this development does not, of necessity, imply that mismatches are increasingly eliminated. It jus means that errors become increasingly elusive, with potential(ly catastrophic) consequences that nobody oversees. You start our accepting small cracks, but cracks have a way of growing as they propagate. (In short, I am inclined to regard all of this as a zero-sum game, with a basic tradeoff between primary thinking and secondary thinking that can never be resolved).

[palindrome]

i just wrote an android app in one afternoon. for me this was completely unthinkable a year ago. i had never in my life written a line of code and now i am writing homepages and mutilingual chatbots powered either by one of the popular llms or a custom AI that i host on my own GPU server. i create php files and work with databases. i just wrote a custom gui for ollama because i hated openwebui. at this time it feels like the only border is my imagination. and i am only a bebop guitarist who up to very recently thought json was just a butchered first name....

This is nice for you, but you also need to consider (as just _one_ example) thousands of legit medical journals in the lower range of impact factors where authors can now do their own statistics without even caring to consult a statistician, and with reviewers unable to tell the difference.

[palindrome]

I'm a real spoilsport. Need to take a break.

[StefanoGhirardo]

A slight tangent, but this essay by Italo Calvino from - I think - the 80s, seems pretty on-point right now.

https://www.jfki.fu-berlin.de/academ...li_reader2.pdf


There's a lot of subtlety in there and, despite now having read it 3-4 times, I'm sure I've missed loads. But one of the points I think he's making is that the onus is on the reader, not the author, to imbue a work of literature with value. He talks about how authors (and machines) can combine words in new patterns, but it's the subconscious of the reader that determines the significance of those patterns (he uses the metaphor of puns as an example), and that authors come up with interesting combinations often by chance, just as we imagine machines would. That being said, I don't think he accounts for the fact that authors, even if these combinations are discovered by chance, will filter and vet them before committing them to the page or the final draft. In other words, they themselves will act as surrogate readers. Machines obviously can't do that. But it's very possible I missed something here. There's then the question I guess of how well an author's subconscious response of a new combination corresponds to that of their readers', determining the success of the phrase.

At any rate, the essay is well worth a read, and I'd love to hear any responses folks have to it, either as it stands, or within the current discussion of 'AI'.

Thanks! An interesting point of view (and a great look forward, from 1967!), and replacing "words", bricks of the literature, with "musical notes" makes us think even more. I found here (Cibernetica e fantasmi: Calvino e l'intelligenza artificiale Calvino) a nice gloss. It's in Italian but I think that, using a bit of AI, it can be translated well, right?

[djg]

authors can now do their own statistics without even caring to consult a statistician, and with reviewers unable to tell the difference.

and this is different from what they used to do without AI in what way exactly?

[palindrome]

and this is different from what they used to do without AI in what way exactly?

Sorry, I didn't really mean to be impolite.

Well, I guess I know from experience that a great many authors publishing at this level (low impact but legit) know next to nothing about statistics (i.e. even less than I do), very much like, BTW, any statisticians involved don't necessarily grasp the full dimensions of the paper's content (i.e. less so than I do).

Higher-ranking journals tend to add statistical advisors to the peer-review process, so authors need to be able to defend their statistical approach. Lower-ranking journals do not routinely do this, so the chances are very real that authors come up with superficiailly plausible analyses using AI without actually understanding them and nevertheless get away with this because the reviewers are no better.

PS edit: These things are far from black or white, but lighter shades of gray are certainly preferable to darker ones.

[Flowboy]

It has recently been the case that following auditing, large numbers of (I think) mostly health related papers, have been withdrawn or rather been forced to, due to unverified claims, poor statistical analysis, or deliberate misrepresentation, or complete failure of peer reviews. Were AI assistants used in the paper writing? I have no idea, but wouldn't be surprised.

As has no doubt been said, (I haven't read the entire thread) AI just hoovers up what is on the net & presents it convincingly. It should be obvious to any user that potentially a large chunk of that data is false, untrue, misrepresented etc. As if you went to the library for some information & were presented with both technical texts & children's comics, (& angry forum posts) without being able to spot the difference, other than by applying your personal knowledge & experience. Easier if you have decent knowledge already - such as the OP knowing about GG's guitar/s but much harder if you know nothing or very little about the subject.

I recently used Ms "Copilot" to ask about a very obscure German luthier. It could only come up with one piece of info about a completely different person with exactly the same name & wrongly suggested it was the same person. I was able to "converse" with the AI, to suggest or query whether there was a link with another even earlier maker with the same surname. Copilot could not find any info & thanked me for pointing out the existence of the Luthier.
It was of course unfailingly polite.

Quite different from the average forum exchange where many people seem to have no sensitivity filter; they freely make, (sometimes in their very first response to a thread) blunt, insulting, disdainful, haughty, etc. etc. ad nauseum comments, & / or have such thin skins that even the faintest suggestion of an insult (if reading between the lines) causes them to take umbrage.
I prefer to allow a little leeway for the badly written, unintentional etc. like something misheard in a noisy pub. After all we are often responding in a hurry, on a little phone, whilst riding a bicycle...

BTW I am now v interested to read the Calvino text.

[PDeville]

"AI" is a scam. There is no intelligence outside of sentient beings, it's just a machine that is programmed to do a job like a robot arm that does spot welding. In this case the robot arm put the weld in the wrong spot multiple times.

Yes it is. Guaranteed to make the population even more stupid and helpless than it already is. The letters will be attached to everything as
another meaningless marketing acronym.
I made my living as a sytems programmer, and I still am slighly surprised by how easily people are mesmerized by and then addicted to this latest stupid shit.

Yesterday you were able to type a few keywords into a search engine string. But, Today you need "AI"

[omphalopsychos]

Of course, even before Flaubert, people knew stupidity existed, but they understood it somewhat differently: it was considered a simple absence of knowledge, a defect correctable by education. In Flaubert's novels, stupidity is an inseparable dimension of human existence. It accompanies poor Emma throughout her days, to her bed of love and to her deathbed, over which two deadly agélastes, Homais and Bournisien, go on endlessly trading their inanities like a kind of funeral oration. But the most shocking, the most scandalous thing about Flaubert's vision of stupidity is this: Stupidity does not give way to science, technology, modernity, progress; on the contrary, it progresses right along with progress!

-The art of the novel, kundera

[Al Haig]

Yes it is. Guaranteed to make the population even more stupid and helpless than it already is. The letters will be attached to everything as
another meaningless marketing acronym.
I made my living as a sytems programmer, and I still am slighly surprised by how easily people are mesmerized by and then addicted to this latest stupid shit.

Yesterday you were able to type a few keywords into a search engine string. But, Today you need "AI"

Noone's preventing you to hit the library like MacGyver when you need to get educated on a topic.

[darkwaters]

Of course, even before Flaubert, people knew stupidity existed, but they understood it somewhat differently: it was considered a simple absence of knowledge, a defect correctable by education. In Flaubert's novels, stupidity is an inseparable dimension of human existence. It accompanies poor Emma throughout her days, to her bed of love and to her deathbed, over which two deadly agélastes, Homais and Bournisien, go on endlessly trading their inanities like a kind of funeral oration. But the most shocking, the most scandalous thing about Flaubert's vision of stupidity is this: Stupidity does not give way to science, technology, modernity, progress; on the contrary, it progresses right along with progress!

-The art of the novel, kundera

Having worked for a crown corporation for 30 years, I can but wholeheartedly agree.

[palindrome]

A slight tangent, but this essay by Italo Calvino from - I think - the 80s, seems pretty on-point right now.
https://www.jfki.fu-berlin.de/academ...li_reader2.pdf

Now that I have read the entire text, I'm still liking the challenge (dont't get me wrong on this) but I don't really see the connection that you seem to be implying. Perhaps I've just lost sight of it.

I really don't know what to say, neither having enough background knowledge in philology or cultural sciences, nor having ever read anything by Calvino before. I find myself wholeheartedly agreeing with some passages, disagreeing with others, and there is definitely a third category of points I don't feel I have properly understood at all.

Then there is the historical perspective, with a need to look at things through my parents' generation's eyes back in the Eighties. Perhaps I'm wrong, but the Chomsky way of thinking (see page 9) seems to be (much) more relevant and up to date than any of these structuralist or even cybernetic considerations (see page 8), although there are exceptions (e.g. Daniel Everett, who considers himself a structuralist, is nevertheless interesting IMO)

Generally speaking, I would agree to a direct connection between "literature" and "machines" only to the extent that prosody is, of course, a mathematical phenomenon. OTOH, I cannot think of literature as the main yardstick for written language, considering that (cuneiform) writing was invented for much more pedestrian reasons, and the vast majority of today's writing serves a completely different purpose as well.

Rather than "playing on words" to process "unconscious" or "preconscious" "myths," "fables," and "taboos," most of us spend time on writing in a perennial struggle to bridge the gap between our thinking and the outside world. I, for one, do not subscribe to the whole idea of language shaping our thinking (other than peripherally).

[palindrome]

BTW, to counteract thread drift, here's where cuneiform script meets the Telecaster:

"...this script moved from a simple business of drawing pictures of ideas into a method of recording sound. And that is the essence of writing, that you have a set of marks which record the sound of the language with its words and grammar and all the components which somebody else can put on the record player and retrieve the words when they read it. This is a miracoulous matter, and the shift from pictographic use to writing sounds was the only _real_ giant leap man has ever made, apart from the development of the electric guitar in about 1952"

Navigate to 8:50 (I tried posting the current-time link, but this doesn't seem to work):

So true.

[RJVB]

I, for one, do not subscribe to the whole idea of language shaping our thinking (other than peripherally).

So, what do you do when you "are thinking", i.e. how do you do it ... and please do include any additional languages in the consideration

[supersoul]

A slight tangent, but this essay by Italo Calvino from - I think - the 80s, seems pretty on-point right now.

https://www.jfki.fu-berlin.de/academ...li_reader2.pdf


There's a lot of subtlety in there and, despite now having read it 3-4 times, I'm sure I've missed loads. But one of the points I think he's making is that the onus is on the reader, not the author, to imbue a work of literature with value. He talks about how authors (and machines) can combine words in new patterns, but it's the subconscious of the reader that determines the significance of those patterns (he uses the metaphor of puns as an example), and that authors come up with interesting combinations often by chance, just as we imagine machines would. That being said, I don't think he accounts for the fact that authors, even if these combinations are discovered by chance, will filter and vet them before committing them to the page or the final draft. In other words, they themselves will act as surrogate readers. Machines obviously can't do that. But it's very possible I missed something here. There's then the question I guess of how well an author's subconscious response of a new combination corresponds to that of their readers', determining the success of the phrase.

At any rate, the essay is well worth a read, and I'd love to hear any responses folks have to it, either as it stands, or within the current discussion of 'AI'.

Thanks for posting this, I hadn't really thought of AI in terms of literary/critical theory (not that I'm well-versed at this type of thinking anyway).
The essay made me think of Roland Barthes' "The Death of the Author" essay (https://sites.tufts.edu/english292b/...the-Author.pdf), which is more about separating the (human) author from a literary work. Destroying the grand narrative, how the reader completes work and actually infuses it with meaning, that the literary work transcends the author's intended meaning. It's all mixed up in my head with general art theory, critical theory, structural and post-structural thinking, and when I think about that stuff too much I just get a headache and/or become sleepy.
Italo Calvino was a member of Oulipo, I think, or at least he was associated with them. Oulipo is interesting to me, because they try to use math to construct literary works, but they are coming at it from a very 60's perspective, where anything can mean anything if you just open your mind to it. Meaning is slippery, and that's a good thing, in this type of thinking. I'm not sure I've ever agreed.
I was/am interested in Oulipo's principles as they relate to music, namely using algorithmic techniques to generate... well, sound in general. At one point I was quite into to different forms of noise and I could use them as flavors. Iannis Xenakis, for instance, had several ideas about stochastic oscillators; the one I explored a lot was called GenDyn, where you could control the randomness of different aspects of a wave form; how much randomness was in the width of a periodic wave, for instance. That has a particular sound.
Right now, with AI, I feel like this has become the endgame, lowest-common demoninator: analyze popular pop songs, find common patterns, use that to generate new, similar-sounding pop songs, introduce it to the population with algorithms designed to attract interest, then make lots of money. Meanwhile, music gets shittier.
I'm still interested in using algorithms and analysis, but I'm still mostly inspired by the American composers from the 60s/70s (Gordon Mumma, Alvin Lucier, David Tudor, David Behrman...) who were inspired by Cybernetics and feedback to get inside the electronic circuitry; artistically engaging with electronics.
Ok, that drifted a lot from whatever original point I had. Well, it's Sunday afternoon and my soft-tissue brain is only half firing correctly thru the mush.

[palindrome]

So, what do you do when you "are thinking", i.e. how do you do it ... and please do include any additional languages in the consideration

Oh, I'm running out of time for today, but I figure these two article might keep you busy while I'm gone:

Language of thought hypothesis - Wikipedia

Linguistic relativity - Wikipedia

[palindrome]

So, what do you do when you "are thinking", i.e. how do you do it ... and please do include any additional languages in the consideration

Actually, one of the references in these Wikipedia articles offers a very nice summary:

Economist Debates: Language: Statements

To answer your question, the last sentence in this overview of pros and cons strikes a tonic chord with me:

"But the amiable idea that language shapes thought has become disconnected, in our popular culture, from any consideration of mere fact; and as a result, nearly every instance of this idea in the mass media is false or seriously misleading"

And, as for the second part of your question, I would add to this: the idea that speaking more than one language would somehow endow a person with multiple identities is apparently so sexy as to get irrationally overblown all the time.

The language-of-the-mind ("mentalese") concept appears plausible enough to me. I'm not saying that my thinking doesn't involve chunks of "public language" (nicely put in one of these Wikipedia articles), but my core thought processes most certainly don't work along these lines, which also means that my differentiation between any distinct "public languages" dissolves into thin air at this level.

Isn't this how you experience your own thinking? (I'm sure it is.)

[CliffR]

Actually, one of the references in these Wikipedia articles offers a very nice summary:

Economist Debates: Language: Statements

To answer your question, the last sentence in this overview of pros and cons strikes a tonic chord with me:

"But the amiable idea that language shapes thought has become disconnected, in our popular culture, from any consideration of mere fact; and as a result, nearly every instance of this idea in the mass media is false or seriously misleading"

And, as for the second part of your question, I would add to this: the idea that speaking more than one language would somehow endow a person with multiple identities is apparently so sexy as to get irrationally overblown all the time.

The language-of-the-mind ("mentalese") concept appears plausible enough to me. I'm not saying that my thinking doesn't involve chunks of "public language" (nicely put in one of these Wikipedia articles), but my core thought processes most certainly don't work along these lines, which also means that my differentiation between any distinct "public languages" dissolves into thin air at this level.

Isn't this how you experience your own thinking? (I'm sure it is.)

Have you read Ted Chiang's The Story Of Your Life or seen Arrival, the movie based on it? It's kind of an extreme take on the Sapir-Whorf hypothesis.

[palindrome]

Have you read Ted Chiang's The Story Of Your Life or seen Arrival, the movie based on it? It's kind of an extreme take on the Sapir-Whorf hypothesis.

Just watched the trailer on Youtube, and no, didn't ring a bell.

So much the better, as "extreme take on the Sapir-Whorf hypothesis" doesn't sound like a good recipe for relaxed entertainment as far as I am concerned.

[CliffR]

Just watched the trailer on Youtube, and no, didn't ring a bell.

So much the better, as "extreme take on the Sapir-Whorf hypothesis" doesn't sound like a good recipe for relaxed entertainment as far as I am concerned.

Lol - well, you have to take it with a pinch of salt. But if you do, it's one of the best SF movies of recent times I would say. It's SF in the spirit of 'what if this idea were true', and following the consequences, which I tend to prefer to having things be strictly physically accurate.