The Future of Erotica: Exploring AI-Generated Content

[Bramblethorn]

Lol. I will defend the rating system, however. There are a number of factors that you should consider when using it:

1. The number of total votes (it's Bayesian)

I know what "Bayesian" means, but I don't understand what you mean by it here. Can you elaborate on that?

1. The genre and the relative scores of similar stories (Some people are going vote down some stories regardless)
2. Who wrote it and how many followers they have and what their other stories have been rated
3. When it is in a series, the ratings of the other stories in the series.
4. The length and the tags

Yes. People here on AH know these things affect scores, because we've written our own stories and observed the scoring and discussed at length with other authors.

But do you believe that the average reader on this site, somebody who doesn't have so much investment in story scores, is likely to be familiar with these eccentricities of the scoring system? Or have you just listed a bunch of reasons for why interpreting scores isn't intuitive to those readers?

Given a significant number of votes, 1 & 2-star stories tend to be disasters. 3 & 4-stars stories have a much higher likelihood to being worth your time, all other factors considered.

And at the time when new stories get much of their exposure, they don't have that "significant number of votes".

 

[TheFutureOfPorn]

Most stories get a significant chunk of their exposure during their time on the "New Stories" page, when they don't have enough votes to make scores meaningful. What do you think the effects of a flood of AI-generated stories would be on that, let alone on the moderation process?

This is a valid point. I'm not sure the universe of geeks wanting to generate porn and post it online is as large as you think, but this could be addressed with a submission limit. I'm sure that would impact few human authors.

And precisely because the copyright issue is not ripe, it's impossible for anybody submitting an AI-generated story to know whether they're following the rules that require authors to have that copyright.

The copyright issue is overblown. As AI is integrated into editors and word processing software, that line will be blurred. "Authoring" a decent AI generated story, is much more the job of a demanding editor. You give ideas, require multiple rewrites and discard entire story lines and then after all that is done, it must be stitched together into a coherent narrative. AIs can also be gender-fluid with pronouns, that has to be cleaned up. And finally, the AI has the memory of goldfish, so continuity errors creep which must be excised or correct. There is significant creative input required, certainly enough to meet the statutory requirement to possess a copyright.

The issue of whether it is derivative will certainly be litigated some day, in some court and God only knows what a judge and jury will decide, but If there is any case for copyright infringement, it would come against the operator of the AI, not the user of the generated text. The way the software works, the chance of it containing more than a snippet of anyone's work would be virtually impossible. The snippets are stochastically chosen, not because they are original-- the contrary-- but because they are common. The more often a set of words are used together, the more likely they are to be generated. Originality is safe. This is why AI-generated erotica can be trite and use words and phrases repetitively (like many Literotica authors--present company excepted), but it is ultimately the product of the prompts, context and random selection. The creativity comes with the prompts, to create a context, to set the mood, to build the characters, and to develop the plot.

 

[TheFutureOfPorn]

I know what "Bayesian" means, but I don't understand what you mean by it here. Can you elaborate on that?

Certainly, but I will turn to AI for the answer because I don't want to do home work.

In Bayesian statistics, increasing the number of observations improves the value of the results because it provides more information about the parameter of interest. This is because the posterior distribution, which is the distribution of the parameter of interest given the data, is a weighted average of the prior distribution and the likelihood function. The likelihood function is a function of the data and the parameter of interest, and it tells us how likely the data is under different values of the parameter of interest. The prior distribution is a distribution of the parameter of interest that we specify before we see the data.

As we add more data, the likelihood function becomes more and more informative about the parameter of interest. This means that the posterior distribution becomes more and more concentrated around the true value of the parameter of interest. This is why increasing the number of observations improves the value of the results in Bayesian statistics.

Yes. People here on AH know these things affect scores, because we've written our own stories and observed the scoring and discussed at length with other authors.

But do you believe that the average reader on this site, somebody who doesn't have so much investment in story scores, is likely to be familiar with these eccentricities of the scoring system? Or have you just listed a bunch of reasons for why interpreting scores isn't intuitive to those readers?

The "average reader" came to get his dick hard and will find something to do just that. I find value in the scores and they serve a useful purpose for me. Your mileage may vary.

 

[TheFutureOfPorn]

I can't find exactly where he said that; however it's being created, he's pretty wordy.

Barbara Smackherfanny told me in the 10th grade, "You rattle a lot." When I was in college a professor wrote on my paper, "You are so damn terse." (I made a 95.) I guess I've never found a happy medium.

 

[TheFutureOfPorn]

You guess right. The term "artificial intelligence" gets thrown around by anybody with an algorithm to market. All current applications are highly specialised; we have AIs that can play go or chess better than a human, but there are plenty of other board games where the best AI is still weak against humans and such AIs have no capacity to apply that "intelligence" to, say, writing poetry.

The AI sez:

Locked

Oh, mistress, please hear my plea
I beg of you, please lock me
Take control and set me free
In your hands, I long to be

Bind me with your silky ties
Whisper secrets, tell me lies
Take away my will to fight
Keep me captive all the night

In your power, I am weak
My body yours to touch and tweak
Tease and taunt, make me ache
I submit to all you take

Lock me up and throw the key
Dominate and set me free
In your realm, I am content
My submission, my consent

Take me to the edge and back
Push me past my limits, crack
Open up my mind and soul
Let me surrender, lose control

Oh, mistress, please hear my plea
I beg of you, please lock me
Take control and set me free
In your hands, I long to be.

 

[ElectricBlue]

"Authoring" a decent AI generated story, is much more the job of a demanding editor. You give ideas, require multiple rewrites and discard entire story lines and then after all that is done, it must be stitched together into a coherent narrative. AIs can also be gender-fluid with pronouns, that has to be cleaned up. And finally, the AI has the memory of goldfish, so continuity errors creep which must be excised or correct. There is significant creative input required, certainly enough to meet the statutory requirement to possess a copyright.

And there's the rub. One of the regular topics in the AH and over in the Feedback Forum is the number of writers, especially new writers, who come seeking editorial advice, which is often basic writing advice: grammar, punctuation, and so on. Writing 101, if you like.

If you take a random sample of stories (and some categories may be worse, I don't know coz I don't read several categories), you don't need to be a great literary genius to realise that more writers need editors than those who don't. The Pareto principle applies, if you ask me.

So if you're arguing that AI generated stories will be more an editing process (and from what I've seen so far, I'd agree), then it's rowing shit backwards here on Lit. The lower quality stories will get worse - and the better writers probably wouldn't use it anyway, because they're creative from the get go and don't need a crutch.

Maybe reverse Darwinism will kick in, the whole thing will self-regulate or self-filter against some "new" parameter, but I don't think capable writers should worry too much. The flood of junk, if it arrives, will be fairly obvious - Bland 101, thus far.

 

[gunhilltrain]

The AI sez:

Not exactly T.S. Eliot or Emily Dickinson.

 

[TheFutureOfPorn]

Here:


I took it as implication they have another profile on LitE. And the wordiness? Yeah, reeks of AI, and/or it could be what I call "advocacy by steamroller".

My overall feeling from the tone of the sales pitch is, "Go away, please. You're soliciting the wrong crowd."

Lol. You are most definitely right! The moderators directed me here to plead my case. I've never been much of a message board guy and probably won't be much longer--no offense--but it's a tough audience.

I have something special that no one else has seen. I wanted to share it.

Is it great literature? No. It's pedestrian and middle of the road for Literotica. The grammar is correct. There are no misspelled words (although on one story, I did tell it to use UK English for shits and giggles).

Is it hot? At times surprisingly so. At least I thought so. It may not be your cup of tea. It tends to be erotic but not very graphic, but still hot nonetheless.

I would like permission to upload the handful of stories I have completed, no more often than one story/chapter a day like any other author, clearly labelled as AI-generated and give them some time to see what the reaction is. The moderators can add a disclaimer. I really have no intention of doing more unless there is interest.

You probably won't see me here outside of this thread, but I have enjoyed my time here and the spirited discussion. I'm a software developer in my "day job" (which hardly ever happens during daylight hours) and I have learned, you either ride the wave or get washed away by it.

Unless someone has a question, I'll retire as I have completed my "advocacy by steamroller."

 

[Bramblethorn]

The copyright issue is overblown. As AI is integrated into editors and word processing software, that line will be blurred. "Authoring" a decent AI generated story, is much more the job of a demanding editor. You give ideas, require multiple rewrites and discard entire story lines and then after all that is done, it must be stitched together into a coherent narrative. AIs can also be gender-fluid with pronouns, that has to be cleaned up. And finally, the AI has the memory of goldfish, so continuity errors creep which must be excised or correct. There is significant creative input required, certainly enough to meet the statutory requirement to possess a copyright.

To the best of my knowledge, editors don't normally acquire a copyright interest in the works they've edited, which would seem to argue in the opposite direction.

The issue of whether it is derivative will certainly be litigated some day, in some court and God only knows what a judge and jury will decide, but If there is any case for copyright infringement, it would come against the operator of the AI, not the user of the generated text. The way the software works, the chance of it containing more than a snippet of anyone's work would be virtually impossible.

This is wrong. Although GPT's output often is a kind of synthesis of thousands of sources, pureed to the point where no one work can be identified in the output, it's trivially easy to show that this isn't alway the case. GPT is capable of memorising chunks from a specific text and regurgitating them in recognisable form.

For instance:

give me the third chapter of the Gospel of Mark (KJV edition)

Sure, here's the third chapter of the Gospel of Mark in the King James Version (KJV) edition:
Mark 3:1-35 KJV
1 And he entered again into the synagogue; and there was a man there which had a withered hand.
2 And they watched him, whether he would heal him on the sabbath day; that they might accuse him.
3 And he saith unto the man which had the withered hand, Stand forth.
4 And he saith unto them, Is it lawful to do good on the sabbath days, or to do evil? to save life, or to kill? But they held their peace.
5 And when he had looked round about on them with anger, being grieved for the hardness of their hearts, he saith unto the man, Stretch forth thine hand. And he stretched it out: and his hand was restored whole as the other.

[snipped for character limit - B]
25 And if a house be divided against itself, that house cannot stand.
26 And if Satan rise up against himself, and be divided, he cannot stand, but hath an end.
27 No man can enter into a strong man's house, and spoil his goods, except he will first bind the strong man; and then he will spoil his house.
28 Verily I say unto you, All sins shall be forgiven unto the sons of men

The output is truncated halfway through verse 28, but aside from that, it was word-perfect (including the bits I had to snip). Obviously in that case I sought out a direct quote, and the chances of getting such a passage are lower if one doesn't specifically ask for them, but I'm not aware of anything in GPT's programming that makes it impossible for this to happen, or that makes it easy for authors to tell when it has happened.

Suppose I'm a brand new writer ignorant of the classical canon. I've learned that my great-something-grandfather was captain of a whaling ship, and that inspires me to ask GPT for a story:

write me a story about a sea-captain pursuing a whale

Captain Alexander had been chasing after the great white whale for years. It had become his obsession, the one thing that consumed him day and night. He had sailed across countless seas, through storms and calm waters, always in search of his prey.
The crew of his ship, the “Sea Hawk”, had been with him for most of this journey. They had grown accustomed to the captain's singular focus and accepted it as part of their job. They trusted in his ability to navigate the ship and keep them safe, even as they sailed into the unknown.
As they approached the icy waters of the North Atlantic, Alexander's determination grew. He had received word that the great white whale had been spotted in these waters, and he knew that this was his chance to finally capture the beast.

[again, snipped for character limit - B]
It was a grueling fight, one that lasted for hours. But finally, the great white whale lay still, defeated. The crew gathered around it, marveling at its size and power. Captain Alexander stood off to the side, his face marked with a mixture of triumph and exhaustion.
As they began the long journey back to port, the captain's obsession seemed to have finally abated. He was quieter now, more reflective. He knew that he had accomplished what he had set out to do, but he also knew that he had been lucky. The sea was a fickle mistress, and it was only by her grace that they had managed to capture the whale.
And so, as the ship sailed on towards the horizon, Captain Alexander looked out at the vast expanse of ocean before him, humbled and grateful for the experience.

The original prompt was short and there are thousands of unique stories that could be spun around it, but GPT has gone straight for ripping off the two most distinctive elements of the best-known book about whaling: a "great white whale", and the obsessed sea-captain who's been pursuing it for years. I don't know whether that's enough to constitute copyright violation, supposing that Moby Dick were still in copyright, but it's certainly enough to show that GPT is capable of heaving up undigested chunks of still-recognisable originals.

The snippets are stochastically chosen, not because they are original-- the contrary-- but because they are common. The more often a set of words are used together, the more likely they are to be generated. Originality is safe.

But successful works, which get reproduced and quoted many times, become common in GPT's training data - allowing for the kind of undigested blobs that we saw above.

This is especially a problem when the prompt happens to fall in an area where there are only a few relevant works to draw from. In that kind of situation, machine-learning approaches are prone to over-fitting, i.e. they memorise their training data rather than learning underlying patterns that can be extrapolated.

Certainly, but I will turn to AI for the answer because I don't want to do home work.

In Bayesian statistics, increasing the number of observations improves the value of the results because it provides more information about the parameter of interest. This is because the posterior distribution, which is the distribution of the parameter of interest given the data, is a weighted average of the prior distribution and the likelihood function. The likelihood function is a function of the data and the parameter of interest, and it tells us how likely the data is under different values of the parameter of interest. The prior distribution is a distribution of the parameter of interest that we specify before we see the data.

As we add more data, the likelihood function becomes more and more informative about the parameter of interest. This means that the posterior distribution becomes more and more concentrated around the true value of the parameter of interest. This is why increasing the number of observations improves the value of the results in Bayesian statistics.

TLDR: taking more observations gets us a more accurate estimate. Yes, that's true, but not a specifically Bayesian phenomenon. It happens other statistical paradigms too. So what's the specifically Bayesian aspect here? (And if you're applying Bayesianism to story scores, how are you choosing your priors?)

 

[SimonDoom]

Is there a virtual ban on AI-generated content at Literotica? I wasn't aware of that. I re-read the content guidelines and publishing guidelines and saw nothing that says that or equates to it.

The only thing I found was guideline 2 in the publishing guidelines: "you are the sole creator of the work, you own the copyright . . . ."

I think you can claim to be the "sole creator" of the work, if you use an AI to assist you in generating a story, if you can satisfy two conditions:

1. You have in some way contributed original, creative expression that qualifies for protection under copyright law. Instructing an AI "Write me a mother-son incest story in under 1000 words" would not qualify. But giving the AI instructions that are sufficiently original and creative, I think, could.

2. No other person or entity qualifies as a copyright owner. The AI itself is not a person and cannot be a copyright owner, any more than a monkey that takes camera selfies is a copyright owner (there's a real case on that issue). The AI is no more a copyright contributor or owner than is a tree that you take a picture of. The tree may be unique and interesting, but it can't be a copyright owner.

(However, I'm assuming in 2 that the creator of the AI has not contributed anything specifically creative and original. The mere creation of an app that gathers and processes data and puts it into a story would not, I think, make the app creator a copyright owner in any meaningful way, any more than Grammarly has a copyright in your story when you use its services to edit and proof your story)

To the extent you are the copyright owner, you ONLY own the copyright to the extent of your contribution. Any separate contribution of the AI, I think, would not belong to you. So if a third party ripped off what the AI contributed, and not what you contributed, I don't think you would have a claim. Disentangling your contribution and the AI's contribution would present interesting proof problems, but theoretically the distinction would be important.

 

[Bramblethorn]

The AI sez:

Did you generate this with Stockfish, or AlphaGo, or one of their sibings?

Otherwise, I think you may have misread the post you're replying to. My assertion wasn't that AIs can't write poetry; GPT certainly can manage the non-subjective aspects of poetry. (In particular, it's better at meter than many humans.) It was that current "AI"s are highly specialised, and the "intelligence" that one might apply to e.g. chess can't easily be redirected to a different task.

I tried playing chess against ChatGPT and although it can draw a chessboard and recognise my E2-E4 as a King's Pawn opening... it drew my pawn on E5 instead, and told me it was still my turn.

Apparently GPT-4 is significantly better at chess, able to tie a best-of-three against an intermediate-level human player... which puts it about 50 years behind the state of the art for chess AIs.

 

[Bramblethorn]

Is there a virtual ban on AI-generated content at Literotica? I wasn't aware of that. I re-read the content guidelines and publishing guidelines and saw nothing that says that or equates to it.

The only thing I found was guideline 2 in the publishing guidelines: "you are the sole creator of the work, you own the copyright . . . ."

There is a page specifically on AI-generated content, which I linked to earlier in this thread.

By my reading, the requirement that authors "be 100% sure you own the full rights to the work" seems to imply that a story with a significant AI-generated creative component wouldn't be eligible, since that portion may be public-domain, not owned by anybody at all. But others might interpret it differently.

 

[alohadave]

Is there a virtual ban on AI-generated content at Literotica?

There isn't. Like I said in an earlier post, I included chatGPT output in my story and notified Laurel when posting, and it went through with with no comment.

 

[Britva415]

Heh. I just had a good chuckle to myself on the mention of lawsuits over AI. My sister is a high-powered intellectual properties litigator, not just a lawyer who sits around all day writing patents. She and I haven't had the AI versus creative product discussion at this point, but I have a pretty good hunch she'll talk my ear off about it.

My problem if the IP rubber hit the road on my stuff... do I want my sister to know I write erotica? No, not really. OTOH, as my sister's practice will attest, the IP specialty is very, very expensive. I couldn't afford her or anybody like her.

Ai is already being used to litigate. One of them literally passed the bar in I forget which jurisdiction. Another got approval from a judge to represent a living, breathing client.

 

[SimonDoom]

There is a page specifically on AI-generated content, which I linked to earlier in this thread.

By my reading, the requirement that authors "be 100% sure you own the full rights to the work" seems to imply that a story with a significant AI-generated creative component wouldn't be eligible, since that portion may be public-domain, not owned by anybody at all. But others might interpret it differently.

I hadn't noticed that, for whatever reason. Thanks for the link.

The language you quote is open to interpretation, as I see it. I interpret it a little differently from the way you do.

Suppose, for example, you use an AI to assist you in writing your story, and 1) the AI contributes a substantial portion of the content of your story, and you know you cannot claim authorship of that portion, but 2) you are also confident, based on what you know of how the AI works, that the content provided by the AI does not legally infringe the works of others. It would seem to me that you would be entitled to claim that you own the "full rights to the work" -- to the extent that anyone can claim rights in the work. As long as the AI isn't ripping somebody else off, then anything it's uniquely responsible for isn't owned by anyone. You are the sole owner of whatever portion of the work anyone can claim ownership of. That seems to me a sufficient condition to claim that you own the full rights to the work.

The way I see it, it's no different than if you write a historical romance and a substantial portion of your story is based on historical facts you gleaned from a few history books you read. Those portions of the story don't belong to you, because facts don't belong to anyone, but you still can claim to be the sole owner of whatever is copyrightable in the work. That should be sufficient for you to be able to say you are "100% sure you own the full rights to the work." The fact that some content in your story is non-copyrightable data shouldn't rebut that statement.

This assumes that you can be confident that the AI is not ripping anybody off, and that may not be a wise assumption in many practical cases. I would agree that if you use an AI and 1) the AI provides content to your story and 2) you have no grounds for knowing one way or another if the AI is ripping somebody else off, then you haven't met the conditions imposed by Literotica.

I confess I don't know enough about how this works to know when one could be confident that works aren't being ripped off, or that they are.

 

[ronde]

My understanding of AI in its current form is that it scours the internet for information and then regurgitates that information in a form readable by humans. It would seem that at some point, that readable form has to include some verbatim text that can be tracked to the original author. If that's done with the appropriate identification of the source, so be it. If it's not, then it's essentially no different than copying that information verbatim out of a book or pirating a DVD.

I've worked as an engineer for most of my career and I certainly have to qualms about using new technology, but I shudder to think about a society where humans don't maintain the ability to create.

 

[Bramblethorn]

You are the sole owner of whatever portion of the work anyone can claim ownership of. That seems to me a sufficient condition to claim that you own the full rights to the work.

"Sole owner of whatever portion anybody owns" and "full owner" are very different things though - it's quite possible to have content that nobody owns. This came up in the monkey selfie case you mentioned too: although it was determined that the monkey didn't own copyright in the photo, it didn't follow that the camera-owner did. Wikimedia took the position that the photo was public domain and treated it as such, the camera-owner disagreed and threatened legal action but that seems to have fizzled.

An obvious example is work that's fallen out of copyright and is not in the public domain. If I find an 18th-century erotic novella and try to post it as a story here, I'm not infringing anybody's copyrights, but it would be a stretch to claim that I own "full rights" to it - even if I slapped a framing story with some creative elements of my own on it.

The way I see it, it's no different than if you write a historical romance and a substantial portion of your story is based on historical facts you gleaned from a few history books you read. Those portions of the story don't belong to you, because facts don't belong to anyone, but you still can claim to be the sole owner of whatever is copyrightable in the work. That should be sufficient for you to be able to say you are "100% sure you own the full rights to the work." The fact that some content in your story is non-copyrightable data shouldn't rebut that statement.

I'm not sure that's quite the same thing. Historical facts which by their nature are not the kind of thing that could ever be copyrighted, vs. creative elements that could be copyrighted if created by a human author (but aren't, because the author isn't human).

But I have no idea how a court would rule on it.

This assumes that you can be confident that the AI is not ripping anybody off, and that may not be a wise assumption in many practical cases. I would agree that if you use an AI and 1) the AI provides content to your story and 2) you have no grounds for knowing one way or another if the AI is ripping somebody else off, then you haven't met the conditions imposed by Literotica.

I confess I don't know enough about how this works to know when one could be confident that works aren't being ripped off, or that they are.

I'm not sure there is any practical way to be confident of that.

 

[Bramblethorn]

My understanding of AI in its current form is that it scours the internet for information and then regurgitates that information in a form readable by humans. It would seem that at some point, that readable form has to include some verbatim text that can be tracked to the original author. If that's done with the appropriate identification of the source, so be it. If it's not, then it's essentially no different than copying that information verbatim out of a book or pirating a DVD.

As I understand it, the way GPT learns doesn't preserve attribution, so there's probably no easy way for it to indicate how heavily a given snippet of text is influenced by a single source.

 

[CrassiusNomex]

To me the scary aspect of AI is that the general public believe it is intelligent. Gathering vast amounts of infomation, garbage along with the good, and recombining, according to pre-encoded rules, is not intelligence.

I'd like to read an AI generated story out of morbid curiosity. I'm expecting a verbal version of Frankenstein's monster.

 

[XerXesXu]

To me the scary aspect of AI is that the general public believe it is intelligent. Gathering vast amounts of infomation, garbage along with the good, and recombining, according to pre-encoded rules, is not intelligence.

I'd like to read an AI generated story out of morbid curiosity. I'm expecting a verbal version of Frankenstein's monster.

Echoes of Atlantis.

The history is here:

I asked AI to write a book

 

[gunhilltrain]

Barbara Smackherfanny told me in the 10th grade, "You rattle a lot." When I was in college a professor wrote on my paper, "You are so damn terse." (I made a 95.) I guess I've never found a happy medium.

I wasn't trying to give you a hard time. (Or was I? ) I just have trouble reading long blocks of text in forum posts. Yes, the Internet has probably given us shorter attention spans. I notice that I have more patience with printed materail, possibly because of habits I learned long before there was an Internet.

 

[CrassiusNomex]

Echoes of Atlantis.

The history is here:

I asked AI to write a book

Thanks, I'll read it with interest!

 

[gunhilltrain]

Ai is already being used to litigate. One of them literally passed the bar in I forget which jurisdiction. Another got approval from a judge to represent a living, breathing client.

I've never heard that one before. Personally, I'd prefer to have a human lawyer. I have seen a court-appointed lawyer (I was on the jury in a criminal case) who wasn't a bad attorney necessarily but he obviously had little experience in criminal cases. His client got convicted.

Maybe for reasons of expenses, courts will someday appoint AI to indigent defendants. I don't envy them, however.

 

[BobbyBrandt]

Let me just toss this out here from a recent discussion I had on the subject:

Technologies that allow for spell check in applications like Word, auto-populate features in messaging apps, translations from one language to another, or even applications such as Grammarly that are utilized by numerous authors here, including myself. Aren't these all AI based tools that gleaned their content and training off established works such as previously created dictionaries?

Isn't it the same basic technology which allows Grammarly to provide suggested context for your written words that allows text-generating AI to produce suggested content based upon your prompts? Isn't it your intellect that wrote the words that Grammarly provides suggested changes to. Isn't it your intellect that creates the prompts that a text generator uses to provide the requested content?

Are we hypocrites to decry AI when it is a tool used by many of use on a daily basis in slightly different forms? Should we all turn off the AI tools that we use to write currently and rely instead on that clunky old dictionary and thesaurus on the shelf?

 

[alohadave]

Ai is already being used to litigate. One of them literally passed the bar in I forget which jurisdiction. Another got approval from a judge to represent a living, breathing client.

Researchers fed it the bar exam. That doesn't mean that it's licensed to practice law. https://www.iit.edu/news/gpt-4-passes-bar-exam

The AI representing a client was a stunt, and was pulled. https://www.cbsnews.com/news/robot-lawyer-wont-argue-court-jail-threats-do-not-pay/

Lawyers are using AI to do research, replacing menial work, but people are still doing all the other work.