Policy regarding advanced aids (GPT, other AI, et al)

[Todd]

It could be a fun challenge to make an LLM produce anything but garbage. It is possible though but it will take some work.

ok, then let me ask you this, do you believe you, a renowned poetry critic and workshopper would have the ability to explain where the poetry generated by the AI fails?

Thanks

Sure and it's based on how training models work. There's a reason people refer to  obvious AI writing as AI slop. It's clean, precise, well-punctuated and empty. In short, it is too safe and derivitive.

Even in this thought experiment, it would be hard to overcome. I'm confident that any solid poet could edit an AI poem into shape. They could impose their style on it to make it live.

That though, isn't that interesting to me. I'd be interested to see if by imposing a prompt upfront could something be created that we would consider good without editing on a first pass.

That's how I'm approaching this discussion.

[milo]

It could be a fun challenge to make an LLM produce anything but garbage. It is possible though but it will take some work.

ok, then let me ask you this, do you believe you, a renowned poetry critic and workshopper would have the ability to explain where the poetry generated by the AI fails?

Thanks

Sure and it's based on how training models work. There's a reason people refer to  obvious AI writing as AI slop. It's clean, precise, well-punctuated and empty. In short, it is too safe and derivitive.

Even in this thought experiment, it would be hard to overcome. I'm confident that any solid poet could edit an AI poem into shape. They could impose their style on it to make it live.

That though, isn't that interesting to me. I'd be interested to see if by imposing a prompt upfront could something be created that we would consider good without editing on a first pass.

That's how I'm approaching this discussion.

Excellent! That is the nature in which I am proposing the discussion. I would be curious if someone could truly tell blind but I also think that would be dishonest.

Good to read you again, BTW

[dukealien]

(Rejoining after the heavy holidays with some maybe-relevant thoughts on the subject as it has evolved here)

The machine finds things related to each other.  It could even be said that the machine *knows* which things are related to, or even belong with, each other.  Its known peccadillos all result from confusing *likely* with *right* - leg goes with horse, so five-legged horse; Brandeis goes with cite, so phony citation titled as authored by Brandeis.  And poem where everything fits but is worse than flat.

Human (for lack of a better word) genius joins things that are *not* related to each other and finds (or creates) the relationship.  "The force that through the green fuse drives the flower" contains metaphors an AI would reject, after finding them only by random walk.

(This line of reasoning comes, somewhat, from current reading on Kabbalistic inspiration by rearranging letters and words - free association which intentionally eschews logic and scientific principles.  Where did Freud get his chops?)

On the suggestion of prompts which would trip up an AI or cause it to produce something genius could improve, three ideas-- 

"Write a legal brief proving that Marbury v. Madison was wrongly decided."

"Write a sonnet regretting that love is eternal."

"Write a nonsense poem in the style of Lewis Carroll."

The last would be easiest for the AI, with probably the worst output.  The first would be amusing, especially the made-up cites.  And the second, I suggest, might result in something worth working on - the point is, an AI would not have proposed that prompt itself.

[milo]

(Rejoining after the heavy holidays with some maybe-relevant thoughts on the subject as it has evolved here)

The machine finds things related to each other.  It could even be said that the machine *knows* which things are related to, or even belong with, each other.  Its known peccadillos all result from confusing *likely* with *right* - leg goes with horse, so five-legged horse; Brandeis goes with cite, so phony citation titled as authored by Brandeis.  And poem where everything fits but is worse than flat.

Human (for lack of a better word) genius joins things that are *not* related to each other and finds (or creates) the relationship.  "The force that through the green fuse drives the flower" contains metaphors an AI would reject, after finding them only by random walk.

(This line of reasoning comes, somewhat, from current reading on Kabbalistic inspiration by rearranging letters and words - free association which intentionally eschews logic and scientific principles.  Where did Freud get his chops?)

On the suggestion of prompts which would trip up an AI or cause it to produce something genius could improve, three ideas-- 

"Write a legal brief proving that Marbury v. Madison was wrongly decided."

"Write a sonnet regretting that love is eternal."

"Write a nonsense poem in the style of Lewis Carroll."

The last would be easiest for the AI, with probably the worst output.  The first would be amusing, especially the made-up cites.  And the second, I suggest, might result in something worth working on - the point is, an AI would not have proposed that prompt itself.

Probably all true and I am sure you are more familiar than I am with AI as it is something I keep meaning to get to but haven't yet.

My curiosity stems from a lifelong obsession of mine as to the objective quality of poetry.  It is a topic I have debated on multiple sides and on multiple occasions whether it was a discussion of "know your audience" or "what makes a poem a poem?" or even the objective value of criticizing any poetry at all if there is no objective quality.

How does this relate? - you may find yourself asking.

Ai, I would predict, would be anathema to poets.  AI generated content would (perhaps should) be disliked.  But is this dislike due to the source or the product?  I would posit this - if the method of generation determines whether a poem is good than there is no true objective quality to poetry at all.  If a poem is good (great?) then it should be good regardless of who (what) wrote it.

Now, you could fairly say - AI could not generate a good (great) poem and I would not be in a position to argue at this point due to my inexperience with AI generated content.

Thanks

[dukealien]

...

Probably all true and I am sure you are more familiar than I am with AI as it is something I keep meaning to get to but haven't yet.

My curiosity stems from a lifelong obsession of mine as to the objective quality of poetry.  It is a topic I have debated on multiple sides and on multiple occasions whether it was a discussion of "know your audience" or "what makes a poem a poem?" or even the objective value of criticizing any poetry at all if there is no objective quality.

How does this relate? - you may find yourself asking.

Ai, I would predict, would be anathema to poets.  AI generated content would (perhaps should) be disliked.  But is this dislike due to the source or the product?  I would posit this - if the method of generation determines whether a poem is good than there is no true objective quality to poetry at all.  If a poem is good (great?) then it should be good regardless of who (what) wrote it.

Now, you could fairly say - AI could not generate a good (great) poem and I would not be in a position to argue at this point due to my inexperience with AI generated content.

Thanks

It's a modern trope to answer any dislike or manifested discomfort with "What are you afraid of?" or even to name and ascribe a fear without asking.

But in the present case, the question might have some exploratory value.  To ask, "Can an AI write a good poem?" implies that someone, somewhere can evaluate such a poem objectively - if that has meaning - or subjectively.  So let's turn it around, often a useful mode of interrogation, and ask, "Can an AI write valid critique of a poem?"  Can an AI write *good* critique of a poem?  Could an AI recognize quality?

And there, perhaps, is the fear:  If an AI can recognize what is good, it will have become possible to turn to it for approval.

Approval is, after all, why we submit to critique.  To some, it may be a matter of drawing others (the critic) into our personal universe and either enslave them to that extent or confirm its objective reality (depending on how nuts we happen to be).  Trusting a machine to give critique isn't especially crazy.  Seeking a machine's *approval* is right of the chart.

[milo]

...

Probably all true and I am sure you are more familiar than I am with AI as it is something I keep meaning to get to but haven't yet.

My curiosity stems from a lifelong obsession of mine as to the objective quality of poetry.  It is a topic I have debated on multiple sides and on multiple occasions whether it was a discussion of "know your audience" or "what makes a poem a poem?" or even the objective value of criticizing any poetry at all if there is no objective quality.

How does this relate? - you may find yourself asking.

Ai, I would predict, would be anathema to poets.  AI generated content would (perhaps should) be disliked.  But is this dislike due to the source or the product?  I would posit this - if the method of generation determines whether a poem is good than there is no true objective quality to poetry at all.  If a poem is good (great?) then it should be good regardless of who (what) wrote it.

Now, you could fairly say - AI could not generate a good (great) poem and I would not be in a position to argue at this point due to my inexperience with AI generated content.

Thanks

It's a modern trope to answer any dislike or manifested discomfort with "What are you afraid of?" or even to name and ascribe a fear without asking.

But in the present case, the question might have some exploratory value.  To ask, "Can an AI write a good poem?" implies that someone, somewhere can evaluate such a poem objectively - if that has meaning - or subjectively.  So let's turn it around, often a useful mode of interrogation, and ask, "Can an AI write valid critique of a poem?"  Can an AI write *good* critique of a poem?  Could an AI recognize quality?

And there, perhaps, is the fear:  If an AI can recognize what is good, it will have become possible to turn to it for approval.

Approval is, after all, why we submit to critique.  To some, it may be a matter of drawing others (the critic) into our personal universe and either enslave them to that extent or confirm its objective reality (depending on how nuts we happen to be).  Trusting a machine to give critique isn't especially crazy.  Seeking a machine's *approval* is right of the chart.

I would love to say you are incorrect based on the fact that AI ALWAYS tells you the poem is good but I cannot as I am overly familiar with the confirmation bias of poets.  On this very site, how many times have we witnessed members posting poems of very questionable quality followed shortly after (after  helpful members attempting to help them) by claims of publishing rights or the strong recognition of "real" poets.

But the real problem is much deeper than that and you hint at it here.:

AI represents true objectivity.  People trust computers in a way they never can trust humans.  It is only a matter of time before AI is considered the objective truth in what is considered quality in human poetry and I could see how that is a concern.

Still - actual good poetry is rare.  Anyone who has participated in online forums to any degree can attest to that.  As a reader, is it worth the risk to gain access to high quality poems?  I mean, for as close as I am to William Shakespeare, I am as close to a modern computer so how should I differentiate other than I perfer Shakespeare becuase we are members of the same tribe.

Thanks

[dukealien]

Perhaps apropos, just saw a YouTube™ video in which an apparently informed presenter described an AI beyond LLMs.  This one seems to have, for lack of better words, a gestalt formed of its observations rather than a word-frequency table.  As the presenter explained, this new sort of AI thinks to talk where LLMs talk to think.

That is, if I'm interpreting this correctly, the new machine works internally with its own thought processes and picture(s) of the world, then contrives to present them in language tokens; a LLM manipulates tokens (so does not really "think" at all).  The new machine apparently is faster, more accurate, and takes considerably less resources in terms of memory and such.  The presenter said it might be the key to human-form robots (LLMs are much too large and inaccurate to operate a butler, not to mention a careful cleaning-lady).  We may be crossing a line from building-sized UNIVACS of early SciFi to holographic "Positronic" robot brains a la Asimov.

Considering we're already encountering the resource and logic problems of the UNIVACS with LLMs, one can but hope we'll avoid Asimov's Three Laws tangles.

As for poetry, the new machines might take an interest and be amenable to critique.  Though music seems a more direct outlet...

[milo]

I briefly checked ☑️ out all poetry or all poet or whatever it’s called yesterday (I have never participated there but I get bored easily) and they had a sidebar AI that was explaining all of the poems. The explanations were not great but you could see all of the commenters speaking to them as if they were established fact. At some point I am sure even some he authors will start acting as if the AI knows more about it than they do.

My boss is considered an expert at the intersection of AI in the industry. At the same moment time he likes the say, “AI won’t replace people but people using AI will replace people who don’t “

He’s wrong but I don’t tell him as I am dependent on my paycheck. He keeps adding use cases for AI to my standard work.

There is a thick feeling of doom right now due to the inevitability of it all coupled with the unbearable weight of ennui in modern life.