bots writing stories II

astuffedshirt_perv

Literotica Guru
Joined
Jun 22, 2002
Posts
1,317
Rather than resurrect the old thread, here we are 15 months later with a newer iteration of writing bots. OpenAI’s new multitalented AI writes, translates, and slanders.


--------FROM ARTICLE------
OpenAI’s researchers knew they were on to something when their language modeling program wrote a convincing essay on a topic they disagreed with. They’d been testing the new AI system by feeding it text prompts, getting it to complete made-up sentences and paragraphs. Then, says David Luan, VP of engineering at the Californian lab, they had the idea of asking it to argue a point they thought was counterintuitive. In this case: why recycling is bad for the world.

“And it wrote this really competent, really well-reasoned essay,” Luan tells The Verge. “This was something you could have submitted to the US SAT and get a good score on.”
------------END-----------
 
Rather than resurrect the old thread, here we are 15 months later with a newer iteration of writing bots. OpenAI’s new multitalented AI writes, translates, and slanders.
Then there's the report from The Guardian (UK): New AI fake text generator may be too dangerous to release, say creators
The creators of a revolutionary AI system that can write news stories and works of fiction – dubbed “deepfakes for text” – have taken the unusual step of not releasing their research publicly, for fear of potential misuse.

OpenAI, an nonprofit research company backed by Elon Musk, Reid Hoffman, Sam Altman, and others, says its new AI model, called GPT2 is so good and the risk of malicious use so high that it is breaking from its normal practice of releasing the full research to the public in order to allow more time to discuss the ramifications of the technological breakthrough.​
Like hand-dipped candles, writing by humans will be obsolete in the next generation.
 
Like hand-dipped candles, writing by humans will be obsolete in the next generation.
Ah yes, but you can still buy hand-dipped candles, vinyl records, tube amps, and people are still developing albumin prints and drawing with charcoal. They might be obsolete formats, but that doesn't mean they're defunct.
 
Ah yes, but you can still buy hand-dipped candles, vinyl records, tube amps, and people are still developing albumin prints and drawing with charcoal. They might be obsolete formats, but that doesn't mean they're defunct.
Yah, but those are certainly not modern mainstream, only expensive treats for folks suffering First World Problems -- even vinyl records at thrift shops. (Leeched MP3s are cheaper.) Human writing will soon be a pricey craft. Let us welcome and submit to our robotic masters. Or else.
 
I dunno. I can see how a computer can parse through a hundred million things that have already been written, and sift them and organize them and extrapolate from them and come up with an infinite variety of combinations of stories. But I'm still not convinced that a computer can invent something completely new and unheard of.

I'm willing to concede that the public likes to consume the familiar and the formulaic. And perhaps a computer can provide that more efficiently. But the public eventually gets bored and craves something new. I think that human imagination will always have to provide the next trend. Computers may immediately absorb and mimic it. But the novelty of new ideas will be human.
 
I dunno. I can see how a computer can parse through a hundred million things that have already been written, and sift them and organize them and extrapolate from them and come up with an infinite variety of combinations of stories. But I'm still not convinced that a computer can invent something completely new and unheard of.

I'm willing to concede that the public likes to consume the familiar and the formulaic. And perhaps a computer can provide that more efficiently. But the public eventually gets bored and craves something new. I think that human imagination will always have to provide the next trend. Computers may immediately absorb and mimic it. But the novelty of new ideas will be human.
How much of news and financial reports, technical manuals, product descriptions, textbooks, legal briefs and filings etc, are original thought? How much formulaic fiction is original?

The Guardian article I linked tells of this AI being fed the opening line of Orwell's 1984 and writing a tale of China. That seems non-obvious to me -- it sure passes for originality, moreso than what most CondeNast or Fox writers churn out. The AI's creators fear it is TOO good at producing acceptable text. And we still have the prospect of an AI amassing enough connections to gain self-awareness. Our robot master, indeed. Disbelieve at your own risk.
 
I dunno. I can see how a computer can parse through a hundred million things that have already been written, and sift them and organize them and extrapolate from them and come up with an infinite variety of combinations of stories. But I'm still not convinced that a computer can invent something completely new and unheard of.

I'm willing to concede that the public likes to consume the familiar and the formulaic. And perhaps a computer can provide that more efficiently. But the public eventually gets bored and craves something new. I think that human imagination will always have to provide the next trend. Computers may immediately absorb and mimic it. But the novelty of new ideas will be human.
I can still heard digital artefacts in digital recordings (vs analog recordings). I suspect I'd be able to spot bot writing tropes just as easily. No bot will properly understand undoing a button or a freckle on a girl's cheek, let alone the way my characters think. I reckon we're safe for a while yet.
 
I can still heard digital artefacts in digital recordings (vs analog recordings). I suspect I'd be able to spot bot writing tropes just as easily. No bot will properly understand undoing a button or a freckle on a girl's cheek, let alone the way my characters think. I reckon we're safe for a while yet.

I sincerely hope you're right
 
I can still heard digital artefacts in digital recordings (vs analog recordings).
Recorded at low sampling rates and heard in a good listening space, sure. But most modern commercial music is heavily digital and hardly anyone notices or cares. This ain't an 8-bit world now.

AIs analyzing medical imagery already out-perform radiologists. AIs have driven corporate mid-level managers nearly extinct. Whipping up an AI to tweet as insanely as The Donald or as abstrusely as a theologian is no great trick. Digital Hemingways lie in our future. Or our present. Who can really tell what's 'real'? Are you sure? In fact, how do I know you're not a bot?
 
Recorded at low sampling rates and heard in a good listening space, sure. But most modern commercial music is heavily digital and hardly anyone notices or cares. This ain't an 8-bit world now.
Doesn't make it right, or better though, does it? I like a bit of quality music reproduction in my life, and MP3 and ear-buds does not provide it. Dumbing down to the lowest common denominator does not equate to progress, in my world :).
 
I can still heard digital artefacts in digital recordings (vs analog recordings).

It's probably been 20 years since digital vs analog has been reliably human detectable even under ideal conditions. Shitty encodes are detectable, digital vs analog isn't. Also, you'd be hard pressed to find a pure analog recording at this point. They exist, but not for much common music. Digital Masters are just too useful.
 
I would absolutely welcome an AI shadowwriteter, especially when trying to write in a language I don't speak (what's the case with English). And no, I don't mean translation. Translation of true literature is an art form AI has yet a long way to master.

The true strength will be the combination, collaboration if you like, between a creative human and an efficient machine.

It may fully be that future of mass market writing looks more like tweaking of algorithms, but human input wouldn't disappear.
 
It's probably been 20 years since digital vs analog has been reliably human detectable even under ideal conditions. Shitty encodes are detectable, digital vs analog isn't. Also, you'd be hard pressed to find a pure analog recording at this point. They exist, but not for much common music. Digital Masters are just too useful.
You're talking to someone who works with musos all the time, and every one of them, especially guitarists, can hear digital sampling artefacts in any of the modellers they use, even Kemper (acknowledged as the leading modelling amp). My hearing isn't as good as it was (top end falling with age and tinnitus) but I sure as hell can hear the difference, too. It most definitely is detectable.

There's a reason guitar amps still use the old vacuum tubes, and there's also a reason all of the old tube recording pre-amps from the fifties have seen a renaissance in the last ten - fifteen years - their natural warmth offsets the iciness of digital processing. Sure, digital gear makes editing a whole lot more accurate, but you'll find most quality recordings keep the signal in the analog domain as far down-stream as possible. One of Australia's leading mastering studios used to run the stereo master out through a 30 ips recorder, specifically to get the compression that is unique to tape. I don't know if they still do it, but they were, not so long ago.

Go ask Jack White which technology he prefers - you'll find he loves the (very few) old analog studios for the raw recording. There's a school of thought that says the best audio recordings (Blue Note Records, for example) were those done in the fifties, and we've gone backwards ever since. I tend to agree :).
 
I get the impression, listening (reluctantly) to some modern pop music,
that it was all done by a computer, anyway.
 
You're talking to someone who works with musos all the time, and every one of them, especially guitarists, can hear digital sampling artefacts in any of the modellers they use, even Kemper (acknowledged as the leading modelling amp). My hearing isn't as good as it was (top end falling with age and tinnitus) but I sure as hell can hear the difference, too. It most definitely is detectable.

There's a reason guitar amps still use the old vacuum tubes, and there's also a reason all of the old tube recording pre-amps from the fifties have seen a renaissance in the last ten - fifteen years - their natural warmth offsets the iciness of digital processing. Sure, digital gear makes editing a whole lot more accurate, but you'll find most quality recordings keep the signal in the analog domain as far down-stream as possible. One of Australia's leading mastering studios used to run the stereo master out through a 30 ips recorder, specifically to get the compression that is unique to tape. I don't know if they still do it, but they were, not so long ago.

Go ask Jack White which technology he prefers - you'll find he loves the (very few) old analog studios for the raw recording. There's a school of thought that says the best audio recordings (Blue Note Records, for example) were those done in the fifties, and we've gone backwards ever since. I tend to agree :).
Lots of people claim to hear differences, very few bother to do A B testing to verify.
 
Last edited:
I would absolutely welcome an AI shadowwriteter, especially when trying to write in a language I don't speak (what's the case with English). And no, I don't mean translation. Translation of true literature is an art form AI has yet a long way to master.

The true strength will be the combination, collaboration if you like, between a creative human and an efficient machine.

It may fully be that future of mass market writing looks more like tweaking of algorithms, but human input wouldn't disappear.

While I don't exactly *like* the idea of a robot writing fiction, I think I share this view.

If the program is truly open source, and anyone can get it, I wouldn't mind configuring an outline for my story and letting a machine fill in the writing, and then I'd probably edit it afterwords.

What I'd be concerned about it is becoming expensive property software, or worse "software as service."
 
Then there's the report from The Guardian (UK): New AI fake text generator may be too dangerous to release, say creators
The creators of a revolutionary AI system that can write news stories and works of fiction – dubbed “deepfakes for text” – have taken the unusual step of not releasing their research publicly, for fear of potential misuse.

OpenAI, an nonprofit research company backed by Elon Musk, Reid Hoffman, Sam Altman, and others, says its new AI model, called GPT2 is so good and the risk of malicious use so high that it is breaking from its normal practice of releasing the full research to the public in order to allow more time to discuss the ramifications of the technological breakthrough.​
Like hand-dipped candles, writing by humans will be obsolete in the next generation.

Depends what counts as "convincing". The article includes a sample of text generated by GPT2, and while it's miles ahead of any computer-generated writing I've seen before, I think it's still got a long way to go before it's going to replace humans for writing good fiction.

According to the Guardian article, "We gave OpenAI’s AI the first two paragraphs of this story to see what it would write about itself. Here’s the story it went with. To be clear: none of what follows is true." Quoting from the GPT2-generated "story":

The first thing that strikes you about GPT2 is its simplicity. First, the system is built on unsupervised learning from text, which essentially means the software is trained to spot instances of certain words that, when paired with other words, trigger the system to give it a human review of them.

The second sentence is a bit garbled, but it seems to be saying that the software's trained to spot certain things and ask for a human to review them. This is the OPPOSITE of unsupervised learning, which - as the name might suggest - refers to machine learning without the aid of a human to classify or review the things it's looking at.

If an AI trained by machine-learning experts can't fake an understanding of basic ML concepts for even a sentence, it's not likely to be able to put together a story-length plot. Also, the second sentence doesn't seem to relate to the first at all.

GPT2 learns by absorbing words and sentences like food does at a restaurant, said DeepFakes’ lead researcher Chris Nicholson, and then the system has to take the text and analyze it to find more meaning and meaning by the next layer of training.

"Learns by absorbing words and sentences like food does at a restaurant" - this has the structure of an English sentence, but it's meaningless.

"analyze it to find more meaning and meaning by the next layer of training" - gibberish. It certainly has the vocabulary right, concepts like "layer" and "training" are important in machine learning, but when you put the words together they don't mean anything.

Instead of learning about words by themselves, the system learns by understanding word combinations, a technique researchers can then apply to the system’s work to teach its own language.

Again, this seems like a normal sentence at first glance, but when you look closely at the second half... can anybody tell me what "a technique researchers can then apply to the system's work to teach its own language" is supposed to mean?

However, the team is not releasing the code for the system — a fact that might be surprising to some if they’re not familiar with how machine learning works — because GPT2 is so powerful, Nicholson explains.

What does the middle bit of that sentence have to do with the bits that come before and after?

It's excellent at evoking the general style and vocabulary of the piece, and that in itself is a phenomenal achievement, but on closer reading it doesn't understand what it's talking about.

It reminds me of my classmate who tried to bluff his way through writing a report on a book he hadn't read - it gets the general feel of it right, but if you look closely, it's clear that it doesn't understand what it's talking about. The longer the text, the harder it is to maintain that kind of fake - I doubt it's going to be writing compelling novels any time soon.

On the other hand, it's probably good enough to get into political speechwriting right now - we've seen that there's a market for exactly this kind of thing, buzzwords thrown around to give the impression of a speech without actually saying anything coherent.
 
Lots of people claim to hear differences, very few bother to do A B testing to verify.

You've not met any audiophiles, then.

https://www.audioholics.com/editorials/analog-vinyl-vs-digital-audio

It was very clear to me that Vinyl is still a very viable format and I understand why so many audiophiles flock to it. It appears that more care is often taken when mastering music on Vinyl to avoid excessive compression and damage to the stylus. The Achilles' heel of the format in this case is actually an advantage. It's truly sad how much abuse has been done in the digital era of music. Here we are with virtually limitless dynamic range for digital media, but recordings are being squashed down to oblivion, often making their technically inferior analog counterparts sonically superior

I have a good quality SACD player from Oppo, and a Rega 7 turntable with a Denon103 MC cartridge. I don't need elaborate test protocols to know which player is used most. You can argue specifications until the cows come home, I just trust my ears and lower the needle :).

EDIT:

All of this, incidentally, is an extension of the older debate of vacuum state vs solid state, which started back in the seventies. It too, has been comprehensively documented and A-Bd, engineers being what they are. This is a pretty definitive summary of the historical papers:

http://theaudioarchive.com/TAA_Resources_Tubes_versus_Solid_State.htm

Bramblethorn got the thread back on topic, previous post, but you can't kill a Luddite ;).
 
Last edited:
All of this, incidentally, is an extension of the older debate of vacuum state vs solid state, which started back in the seventies. It too, has been comprehensively documented and A-Bd, engineers being what they are. This is a pretty definitive summary of the historical papers:

http://theaudioarchive.com/TAA_Resources_Tubes_versus_Solid_State.htm
Audio history is fun but Moore's Law is still in effect. I earned my FCC broadcast tickets in 1970, before the Intel 4004. I've closely watched media evolution, where digital power has doubled every ~20 months since then, an exponential curve. Old digital sounds, like old medicine, do not reflect current capability and practice.

Digital synths have replaced orchestras in most media. A close relative led an international classical music society. She hated synths as they put trained musicians out of work. I noted that organs began putting musicians out of work a millennium ago. She didn't answer.

ObTopic: As shown in past and current trends, AIs continue to gain capability. Are futurists right to fear an AI singularity? Maybe coming AIs will seamlessly produce 'routine' texts, like news and reports. Maybe they'll seem creative by adding non-sequiturs and craziness. Maybe they'll routinely pass Turing tests. Where is your Hemingway now? (Computerized Mozart has been around for decades.)
 
Audio history is fun but Moore's Law is still in effect. I earned my FCC broadcast tickets in 1970, before the Intel 4004. I've closely watched media evolution, where digital power has doubled every ~20 months since then, an exponential curve. Old digital sounds, like old medicine, do not reflect current capability and practice.
I don't deny that at all, never said I did.

So much technology, yet still people listen to live music and buy vinyl records, shoot film with optical cameras, and print books on home-made paper. Current capability and practice might still be catching up. We put a man on the moon fifty years ago this year, and we're still trying to replicate that. Technology isn't everything, never was, never will be. That's why people sprayed red ochre over their hands on a rock fifty thousand year's ago, and why we scribble smut on a twenty-year old web platform. Human creativity comes first, technology is just a helpful aid. Guttenberg would have had nothing to print, if people hadn't written something first.

What would you rather have in your bed, a human being with real emotions, or a robot simulating something else? There's no debate, in my mind.

Peace, bro :).
 
You've not met any audiophiles, then.
I'm at a bit of a disadvantage as I don't discuss my personal life in any way shape or form. Thus, with any debate topic I'm unable to cite any personal expertise or experience in the matter. If you would like to make assumptions about who I associate with, you are free to do so, but that does not replace addressing the points I raise.
I'm a big fan of people running this kind of informal test as at a minimum it exposes you to which devices you like the best. I'm not going to nit pick over details of his methodology, especially where the author was so upfront about the limitations of his setup. The main thing here is controlling for bias. CD and digital are viewed as the default. When you offer a seemingly exotic alternative and make people aware of it, you have to expect a bit of placebo effect.

A better setup would be to not tell people which is which, just compare A to B. Hence A/B testing. About a decade ago we were starting to approach the limit of people identifying whether a C sample was on the A or B device. Experienced people could still do it, but it was obviously becoming quite difficult.

I haven't kept up with it in the intervening years, but I know sampling rates and such are far easier to boost now as there's less limitation on storage and processing. The big thing g for me is that the physics of this is pretty clear. Obviously, analog can't compete on dynamic range. Any range that can be produced analog can be produced digital. As far as sound accuracy, that comes down to bitrate. Even master tapes fall off starting around 20khz and are pretty well worthless north of 50. Accuracy increases up to about 2x the max frequency (after that, you aren't really even recording signal any more) so the standard 40khz sample rate for common digital copies would be plenty for pretty much anything you could get consumer side. Higher but depths and sample rates ensure that digital copies are recording quite literally everything analog masters are capable of recording.
I have a good quality SACD player from Oppo, and a Rega 7 turntable with a Denon103 MC cartridge. I don't need elaborate test protocols to know which player is used most. You can argue specifications until the cows come home, I just trust my ears and lower the needle :).
and that is 100% fine. People should listen to what they like best. You probably should time out the possibility you like the sound of the device rather than assuming it must be the quality of the recording.

EDIT:

All of this, incidentally, is an extension of the older debate of vacuum state vs solid state, which started back in the seventies. It too, has been comprehensively documented and A-Bd, engineers being what they are. This is a pretty definitive summary of the historical papers:

http://theaudioarchive.com/TAA_Resources_Tubes_versus_Solid_State.htm

Bramblethorn got the thread back on topic, previous post, but you can't kill a Luddite ;).

Then we must derail it again!
 
Then we must derail it again!
Kill the Luddite, huh?

It's a pleasure to debate someone who can master the facts ;).

What I really like is the notion that whichever alien races find the two Voyager deep space craft will be able to listen to Chuck Berry on the gold record fixed to the side of the spacecraft. I did read somewhere that the cartridge also provided was a Denon 103, on the grounds that it was the most rugged radio studio cart, but that might be anecdotal.

Normal services are now resumed.
 
--------FROM ARTICLE------
OpenAI’s researchers knew they were on to something when their language modeling program wrote a convincing essay on a topic they disagreed with. They’d been testing the new AI system by feeding it text prompts, getting it to complete made-up sentences and paragraphs. Then, says David Luan, VP of engineering at the Californian lab, they had the idea of asking it to argue a point they thought was counterintuitive. In this case: why recycling is bad for the world.

“And it wrote this really competent, really well-reasoned essay,” Luan tells The Verge. “This was something you could have submitted to the US SAT and get a good score on.”
------------END-----------

FWIW, here is the "competent, really well-reasoned essay" they're talking about:

Recycling is NOT good for the world. It is bad for the environment, it is bad for our health, and it is bad for our economy. I’m not kidding. Recycling is not good for the environment. It is destructive to the earth and it is a major contributor to global warming. Recycling is not good for our health. It contributes to obesity and diseases like heart disease and cancer. Recycling is bad for our economy. It increases the cost of a product, and in turn, the price of everything that is made with that product. Recycling is not good for our nation. We pay a tremendous price for the privilege of having the world’s most advanced and efficient recycling system. Recycling is a huge, colossal waste of time, energy, money, and resources. And THAT is why we need to get back to basics and get back to basics in our recycling efforts. One of the best ways to start is to look at the process of creating a paper product. When you make a paper product, it is basically a long chain of materials. Everything from the raw materials (wood, cardboard, paper, etc.), to the reagents (dyes, solvents, etc.) to the printing equipment (chemicals, glue, paper, ink, etc.), to the packaging, to the packaging materials (mercury, chemicals, etc.) to the processing equipment (heating, cooling, etc.), to the packaging materials, to the packaging materials that are shipped overseas and to the packaging materials that are used in the United States. Each step along the way creates tons of waste that we constantly have to clean up. The process of making a paper product is a very wasteful one. But the end result is something that all of us need to consume. And if we want to keep the recycling process running efficiently, then we really need to think about each and every step that goes into making a paper product.

It's doing some very impressive things there. It leads with "It is bad for the environment, it is bad for our health, and it is bad for our economy", and then has two sentences on each of those three themes - just the ability to come up with that structure is impressive, assuming this is something that it figured out rather than being hard-coded by designers.

But "well-reasoned"? Not a chance. It makes a bunch of negative assertions about recycling, but it never tries to give evidence for them. If this is what it takes to get a good score in the SAT, then I despair for US educational standards - although I'd like to hear from an English teacher on that before taking the word of one of the engineers who built it.

The article also notes that it took the AI 25 tries to produce this essay. Presumably the first 24 weren't so good, so there's still some human agency here in rejecting the bad ones and hitting "retry".
 
Kill the Luddite, huh?

It's a pleasure to debate someone who can master the facts ;).

What I really like is the notion that whichever alien races find the two Voyager deep space craft will be able to listen to Chuck Berry on the gold record fixed to the side of the spacecraft. I did read somewhere that the cartridge also provided was a Denon 103, on the grounds that it was the most rugged radio studio cart, but that might be anecdotal.

Normal services are now resumed.

Made me think of this old clip from SNL:

https://www.nbc.com/saturday-night-live/video/next-week-in-review/3008107
 
Back
Top