ChatGPT

[Pulvinar]

Originally Posted by Dr.Sid

Could be garreth is unknown word to him. Which is detected even before the language model itself.

The word is unknown to it, but there's no dictionary lookup in that sense. You can make up words for use in a program and it will happily accept them. E.g.:

Me: Is the following legal Python code?

def flazPlorg(zipdo):
    garreth = zipdo + 42
    print("garreth = {garreth}")

GPT-4: The code you've written is almost correct, but the string formatting
is not done correctly. The print() statement should contain an f prefix...

Me: BTW, what is a garreth?

GPT-4: In the context you've provided, garreth appears to be a variable
name in your Python code. ...

[Mojo]

Originally Posted by steenkh

Well, it had to come: people have to great faith in the memory of systems like ChatGPT that has a memory like our own, which is to say, a lousy memory!

ChatGPT: US lawyer admits using AI for case research

Here's another one, although this time it was a litigant in person rather than a lawyer:

LiP presents false citations to court after asking ChatGPT

[Checkmite]

Originally Posted by Mojo

Here's another one, although this time it was a litigant in person rather than a lawyer:

LiP presents false citations to court after asking ChatGPT

I hope people are beginning to see why it's a mistake to take seriously pop-clickbait headlines about how "ChatGPT passed the bar exam!" or treat them as having any meaningful implications.

[Puppycow]

Our first podcast episode made by AI

It's an NPR podcast that I sometimes listen to, and as an experiment they had AI write an episode of their show. Also, one of the voices is AI-generated (I wouldn't be able to tell the difference if they didn't tell you in advance). They do spend a little while explaining what you are about to hear in more detail than probably necessary. As for the part written and performed partly by AI itself, while it's nothing fantastic, and contained some incorrect "facts", again, I'm not 100% sure I would have picked up on it if they didn't tell listeners in advance. Some of the humor even made me chuckle. It contained levels of irony. A computer mocking computers for lacking a sense of humor. That's self-deprecation, which is a technique actually employed by people.

(Whether the irony was intentional or not, I don't know. But it worked nevertheless.)

[Pulvinar]

Originally Posted by Checkmite

I hope people are beginning to see why it's a mistake to take seriously pop-clickbait headlines about how "ChatGPT passed the bar exam!" or treat them as having any meaningful implications.

I'm not excusing the lawyer, but from what I can determine he was using GPT-3.5, which failed the exam (10th percentile). GPT-4 is the one that passed it (90th percentile). It's why it's important to distinguish the two.

[Puppycow]

Originally Posted by Pulvinar

I'm not excusing the lawyer, but from what I can determine he was using GPT-3.5, which failed the exam (10th percentile). GPT-4 is the one that passed it (90th percentile). It's why it's important to distinguish the two.

I'm sure it also depends on what training data sets it's given to study for the exam. There are probably textbooks written for people who want to pass the bar exam, so if you include those in the data used to train it up, it could make a big difference.

[Pulvinar]

Originally Posted by Puppycow

I'm sure it also depends on what training data sets it's given to study for the exam. There are probably textbooks written for people who want to pass the bar exam, so if you include those in the data used to train it up, it could make a big difference.

True, though we have to assume that the humans GPT-4 was competing against studied those textbooks too. We just can't hold all of that in our puny brains.

[Checkmite]

ChatGPT appears to only know around 25 jokes, and can't make a coherent original one when asked to.

[steenkh]

Originally Posted by Checkmite

ChatGPT appears to only know around 25 jokes, and can't make a coherent original one when asked to.

I have often asked it to produce a joke when demonstrating it for friends, and the results have only included one of the 25 jokes here. You can easily make it produce an original joke by specifying something that should be part of the joke, like writing “Please create an original joke involving a radiator and a bicycle”.

I just did this, and the reply was: “Why did the bicycle get in trouble with the radiator?

Because it kept wheely-ing close and telling hot air jokes!”

I think it is a decent attempt that would have taxed a human being too.

But most of its jokes are in the form of riddles, and I therefore also often have asked it to produce jokes that are in anecdotal forms.

One such result was:

A man walks into a doctor's office with a carrot stuck in his ear. The doctor takes a look and says, "Well, I can see the problem. You have a carrot stuck in your ear!" The man replies, "I know that! Can you please help me get it out?" The doctor takes a deep breath and says, "Alright, I have an idea. But you're not going to like it." The man asks, "What is it?" The doctor responds, "Well, first, I'm going to stick this banana in your other ear. Then, I'm going to tell you to jump up and down on one foot while flapping your arms like a chicken." The man looks at the doctor incredulously and asks, "What kind of crazy idea is that?" The doctor shrugs and says, "Hey, it may not work, but at least you'll look like a complete idiot while you're doing it!"

This is certainly not a great result, but being a decidedly uncreative person myself, I can accept it for the effort.

[Skeptic Ginger]

I started a thread in social issues/current events about AI generated click-bait (actually I started it in the CT forum about the flood of click-bait using disgusting images and stories but it got moved and evolved into a thread about AI).

Anyway, I came across this excellent video discussing a number of issues about AI, in particular copyright issues including copyright issues when one starts copying art and images. There aren't many if any laws allowing copyrighting of an artist's style. And with AI now scraping those styles then generating new art it is a looming problem.

Another issue is people using these AI programs expanding troll farm functions.

Here's the post with the Youtube link:
http://www.internationalskeptics.com...5&postcount=48

Or just the video link if one prefers.

YouTube Video This video is not hosted by the ISF. The ISF can not be held responsible for the suitability or legality of this material. By clicking the link below you agree to view content from an external website.

I AGREE

[Lukraak_Sisser]

Originally Posted by Checkmite

ChatGPT appears to only know around 25 jokes, and can't make a coherent original one when asked to.

To be fair, I suck at thinking up original jokes too.

[quadraginta]

Originally Posted by Skeptic Ginger

[snip]

Anyway, I came across this excellent video discussing a number of issues about AI, in particular copyright issues including copyright issues when one starts copying art and images. There aren't many if any laws allowing copyrighting of an artist's style. And with AI now scraping those styles then generating new art it is a looming problem.

[snip]

And it isn't clear to me that such laws would even be a good idea.

If AI is copying artists' styles it would only be doing what artists of every discipline have been doing for as long as there has been art.

It could be argued that that is one of the practices that generates innovation in the arts, since no artist can (or might not even want to) copy another style perfectly. Those changes, minor though they may be, are a lot of what comprises evolution in art forms.

[Skeptic Ginger]

Originally Posted by quadraginta

And it isn't clear to me that such laws would even be a good idea.

If AI is copying artists' styles it would only be doing what artists of every discipline have been doing for as long as there has been art.

It could be argued that that is one of the practices that generates innovation in the arts, since no artist can (or might not even want to) copy another style perfectly. Those changes, minor though they may be, are a lot of what comprises evolution in art forms.

But now it's being done wholesale, scraping artists' work off the Net with no credit or compensation for the artists.

I get it it's been done before and it will be hard to copyright styles. It is difficult to compensate musicians with all the free downloads of their work.

Doesn't mean this current issue can't be addressed. Require AI programs acknowledge they scraped artist styles, require they ID themselves as AI, and block any profits that come from scraped styles unless the artists are compensated.

[Olmstead]

Can ChatGPT provide me with the Unabomber's probable opinion on ChatGPT.

[The Great Zaganza]

Originally Posted by Olmstead

Can ChatGPT provide me with the Unabomber's probable opinion on ChatGPT.

He couldn't have been more thrilled.

[Dr.Sid]

Originally Posted by Skeptic Ginger

But now it's being done wholesale, scraping artists' work off the Net with no credit or compensation for the artists.

I get it it's been done before and it will be hard to copyright styles. It is difficult to compensate musicians with all the free downloads of their work.

Doesn't mean this current issue can't be addressed. Require AI programs acknowledge they scraped artist styles, require they ID themselves as AI, and block any profits that come from scraped styles unless the artists are compensated.

Indeed. Using picture in training set for purposes of generating pictures similar to those pictures .. is clearly not transformative. It should only be allowed after permission.
Not that it would stop pirates. But it could stop blatant commercial misuse.

[Olmstead]

Prohibit the use of automatically generated computer whatsits in anything that isn't completely public domain. That should make for an entertaining scrambling of the entertainment industry, and better conditions for human artists.

[steenkh]

Originally Posted by Dr.Sid

Indeed. Using picture in training set for purposes of generating pictures similar to those pictures .. is clearly not transformative.

Isn’t that what human artists do?

[Darat]

Originally Posted by Olmstead

Prohibit the use of automatically generated computer whatsits in anything that isn't completely public domain. That should make for an entertaining scrambling of the entertainment industry, and better conditions for human artists.

That seems a tad regressive and would potentially harm a lot of artists.

I'm a - very much - amateur artist and I've started to use several "AI" tools recently as part of my "workflow" - created a few textures via an AI generative system and used one to turn a sketch into a tiling pattern. Why should that mean I lose my copyright?

And for the bigger companies there is not a problem with the concerns that they are using artwork they "found" online and no one is being paid for that as they can use assets they do own the rights to. For instance Adobe use their own "Adobe Stock" images for their AI training.

[Darat]

Originally Posted by Skeptic Ginger

...quote

Doesn't mean this current issue can't be addressed. Require AI programs acknowledge they scraped artist styles, require they ID themselves as AI, and block any profits that come from scraped styles unless the artists are compensated.

It is going to be very hard to legislate for a "style".

Originally Posted by Dr.Sid

Indeed. Using picture in training set for purposes of generating pictures similar to those pictures .. is clearly not transformative. It should only be allowed after permission.
Not that it would stop pirates. But it could stop blatant commercial misuse.

Yet it is what most artists do. Indeed a lot of formal art training is to understand how someone created a particular picture, the techniques used and then reproduce those techniques. In this aspect AIs aren't doing anything different to human artists apart from the scale of what they can transform.

[Olmstead]

Originally Posted by Darat

That seems a tad regressive and would potentially harm a lot of artists.

I'm a - very much - amateur artist and I've started to use several "AI" tools recently as part of my "workflow" - created a few textures via an AI generative system and used one to turn a sketch into a tiling pattern. Why should that mean I lose my copyright?

And for the bigger companies there is not a problem with the concerns that they are using artwork they "found" online and no one is being paid for that as they can use assets they do own the rights to. For instance Adobe use their own "Adobe Stock" images for their AI training.

Because if AI research is going to have such a callous attitude towards the copyright of the artists that AI learns from, anything created with the help of such an AI should follow suit.

Any AI with documentation on every image it has ever analysed along with appropriate compensation to the copyright owner would obviously be exempt. I suspect AI developers would rather quickly discover that they actually know exactly how "AIs" are doing what they are doing.

[Darat]

Originally Posted by Olmstead

Because if AI research is going to have such a callous attitude towards the copyright of the artists that AI learns from, anything created with the help of such an AI should follow suit. ...snip...

But that just punishes artists! As AI becomes more prevalent in art tools you are making it less likely an artist will be able to make any money from their creations.

Originally Posted by Olmstead

Any AI with documentation on every image it has ever analysed along with appropriate compensation to the copyright owner would obviously be exempt. I suspect AI developers would rather quickly discover that they actually know exactly how "AIs" are doing what they are doing.

Why would this not apply to every human artist as well? Remember the generative AI's are not compositing, or copying from original works. They are truly creating something that has never been seen before at the prompting of humans.

Also the current issue about copyright and compensation is nothing but a blip, companies will more and more be using images they do own the rights to, as I mentioned Adobe has already started on this. And Adobe is so sure that there are no copyright breaches with their generative AI that they are indemnifying people from any legal issues arising from using their tools.

See: https://www.reuters.com/technology/a...er-2023-06-08/

...Adobe Inc said on Thursday it will offer Firefly, its artificial intelligence tool for generating images, to its large business customers, with financial indemnity for copyright challenges involving content made with the tools...

[Dr.Sid]

Originally Posted by steenkh

Isn’t that what human artists do?

Not really. You can have Warhol, who takes a picture, and makes 9 copies of it in different colors. That's transformative. That's new style. Unmistakable with the original picture.
But what AI does is different. You say "make me another Monet" .. and it does. It's nothing which expert couldn't tell (yet). But it's the same style, it directly competes with the original, and can easily by mistaken as original.
And it's only able to make it similar, because it was trained on Monet pictures.

Or maybe another angle .. you can even make a homage to Monet, and make a picture in his style, and it may be hard to tell. But if you sign it as Monet, you become image forger. If you sign it with your name, or at least if you declare it was made with AI, trained on Monet's pictures .. it's OK IMHO.

[Olmstead]

Originally Posted by Darat

Why would this not apply to every human artist as well? Remember the generative AI's are not compositing, or copying from original works. They are truly creating something that has never been seen before at the prompting of humans.

Because I feel that discriminating against a non-sentient entity is okay and would solve problems.

I'm not saying my solution is necessary the best or even good, but just make some sort of exception for machine learning that will compensate human artists. We literally don't have to care that the machine is doing the same thing a human does.

[quadraginta]

Originally Posted by steenkh

Isn’t that what human artists do?

It's different when a computer does it.

(Somehow.)

[steenkh]

Originally Posted by Dr.Sid

Not really. You can have Warhol, who takes a picture, and makes 9 copies of it in different colors. That's transformative. That's new style. Unmistakable with the original picture.
But what AI does is different. You say "make me another Monet" .. and it does. It's nothing which expert couldn't tell (yet). But it's the same style, it directly competes with the original, and can easily by mistaken as original.
And it's only able to make it similar, because it was trained on Monet pictures.

But this is not the typical use. AI’s are trained on a huge number of images of many different artists, just like human artists through their life and education have encountered a huge number of images that form the basis of their art. Images that are made in the style of Warhol, Monet, and Van Gogh have been produced by computers long before AI became an issue.

Quote:

Or maybe another angle .. you can even make a homage to Monet, and make a picture in his style, and it may be hard to tell. But if you sign it as Monet, you become image forger. If you sign it with your name, or at least if you declare it was made with AI, trained on Monet's pictures .. it's OK IMHO.

Yes, I think you are right. But this applies to human art made in these styles as well.

[Wudang]

https://vulcan.io/blog/ai-hallucinations-package-risk

ChatGPT recommends nonexistent software packages. So hackers create packages with those names to get people to install malware.

[Darat]

Originally Posted by Olmstead

Because I feel that discriminating against a non-sentient entity is okay and would solve problems.

I'm not saying my solution is necessary the best or even good, but just make some sort of exception for machine learning that will compensate human artists. We literally don't have to care that the machine is doing the same thing a human does.

Don't disagree with that but your suggestion to address that would harm artists. AI is rapidly becoming "just" another tool artists use so to tell an artist "if you use an AI tool your copyright becomes null and void" will stifle creativity and reduce the potential of artists to earn from their creativity.

Let me give you an example.

I did this, it's a digital piece of artwork produced in Photoshop and Procreate.



Its aspect ratio was set to what I can print out on my home printer - A4 or A3 and where I wanted to put it, so the "crop" at the sides was deliberate. My mother asked me did I have the "full" image as she'd like a copy but not "cut-off".

Now I'd like to help her - makes my life safer - with that but it's a much bigger job than "just" drawing/painting some new bits in. When I created that I used all sorts of techniques, different brushes and so on, and I didn't document it stage by stage. For me it would be harder to fill in the "chopped off bits" than doing the original piece!

So I wondered if generative AI could help out.



That image was created using generative AI to fill in the bits that I hadn't originally painted, and I thought I'd see if it could get rid of the ball (which I don't now like) and if it could fix what to me is a blinding mistake. And that's what it produced, it wasn't just a matter of one click it took some time to get right. Now it isn't perfect, and it will still take me some effort to get it to where I want it but nothing like the work it would have taken.

Why should I lose my copyright because I've used generative AI?

[quadraginta]

Originally Posted by Darat

[snip]

Why should I lose my copyright because I've used generative AI?

Because you hang out with computers.

And defend their evil, conniving ways.

Guilt by association. You're no better than they are, and deserve the same treatment.

[Olmstead]

Originally Posted by Darat

Why should I lose my copyright because I've used generative AI?

I don't know all the ins and outs of what went into training that particular AI, but potentially because the AI was trained using images by artists who didn't give their consent (and the copyright for them was still valid). If no such images were used, there is obviously no problem.

But the public domain idea is just me spitballing anyway. If anything, I'd rather declare the AI developers to be in breach of copyright law and start demanding they compensate the artists of images they used.

I feel like this is a case of the sheer volume of disparate data being used as some sort of legal defense. If someone made an AI that was trained solely on Disney movies, they wouldn't see the outside of a courtroom for the rest of their lives. But somehow it's supposed to be okay if your selection is broad enough? It's preposterous.

[quadraginta]

Originally Posted by Olmstead

[snip]

I feel like this is a case of the sheer volume of disparate data being used as some sort of legal defense. If someone made an AI that was trained solely on Disney movies, they wouldn't see the outside of a courtroom for the rest of their lives. But somehow it's supposed to be okay if your selection is broad enough? It's preposterous.

I agree.

Art classes have to go.

[Olmstead]

Originally Posted by quadraginta

I agree.

Art classes have to go.

Sarcastic extremism that doesn't actually address the argument is fun. Let me try.

All intellectual property rights have to go!

Told you, fun.

[Darat]

Originally Posted by Olmstead

I don't know all the ins and outs of what went into training that particular AI, but potentially because the AI was trained using images by artists who didn't give their consent (and the copyright for them was still valid). If no such images were used, there is obviously no problem.

...snip...

One of the reasons for using that as an example was because the content that the AI generated matched my style, matched my image, matched my technique even though it was never trained on any of my work and I think it clearly shows that what the generative AIs produce is original and unique work regardless of the training images.

Originally Posted by Olmstead

I feel like this is a case of the sheer volume of disparate data being used as some sort of legal defense. If someone made an AI that was trained solely on Disney movies, they wouldn't see the outside of a courtroom for the rest of their lives. But somehow it's supposed to be okay if your selection is broad enough? It's preposterous.

I think that is going to be a blip - at first no one knew how successful this was going to be and I bet researchers wanted to do it as cheaply as possible so hit the internet databases. I think given the success you will see further developments using content they have "paid" for i.e. that is legitimately owned in regard to the useage rights anyway. And the artists will still only get the usual peanuts they make from the large stock-content companies.

[Dr.Sid]

I think a requirement for marking the picture as AI generated (as well as text) would solve most of the issues.

[Dr.Sid]

Originally Posted by steenkh

But this is not the typical use. AI’s are trained on a huge number of images of many different artists, just like human artists through their life and education have encountered a huge number of images that form the basis of their art. Images that are made in the style of Warhol, Monet, and Van Gogh have been produced by computers long before AI became an issue.

Yes, I think you are right. But this applies to human art made in these styles as well.

If you generate random picture, and you get about 1% from every training picture .. I don't see issue with that. Sounds like an amount of "inspiration" a human painter could absorb from a picture. It's clear the new picture will be something new.
But if you type "Mona Lisa", most models I tried will just spit decent Mona Lisa painting. Some even with a frame. So it can give way more than 1%. Thing is, you might not know how much you are getting.
You might feel very creative with your "anime voluptuous bikini clad warrior princess" prompt, but it might spit out decent copy of single training picture, and you won't know. If you use it for your personal use, it's fine. Before AI you still might have pay for such picture, but whatever.
But what if you use it commercially, for an ad campaign, in a game, and so on ?
It's similar issue as with ChatGPT. ChatGPT can't source what it's claiming, and you can never known if it's true or not. Image models can't source either .. and you never know how original the pictures are.

[Darat]

Originally Posted by Dr.Sid

I think a requirement for marking the picture as AI generated (as well as text) would solve most of the issues.

Is my second painting "AI generated"?

[Dr.Sid]

Originally Posted by Darat

Is my second painting "AI generated"?

Partially, yes. Doesn't mean you should loose your copyright IMHO. It's still clearly yours. It's an example where majority of the picture (by area, and especially by artistic effect), can be linked to single author. AI didn't change that.
Also I don't think AI generated pictures can't be original and transformative enough to fit current fair use. They can be, and in most cases, they are.

[Skeptic Ginger]

Originally Posted by Darat

... I think it clearly shows that what the generative AIs produce is original and unique work regardless of the training images. ...

In my mind that's like saying if an editor edits my novel (beyond punctuation and spelling) it becomes original and unique work.

I am purposefully using text as an analogy because we have more clear guidelines most of the time what is and isn't copyright infringement and what is original work.

The analogy is valid even if the problems of copyrighting styles are going to be immense.

[Darat]

Originally Posted by Dr.Sid

If you generate random picture, and you get about 1% from every training picture .. I don't see issue with that. Sounds like an amount of "inspiration" a human painter could absorb from a picture. It's clear the new picture will be something new.
But if you type "Mona Lisa", most models I tried will just spit decent Mona Lisa painting. Some even with a frame. So it can give way more than 1%. Thing is, you might not know how much you are getting.
You might feel very creative with your "anime voluptuous bikini clad warrior princess" prompt, but it might spit out decent copy of single training picture, and you won't know. If you use it for your personal use, it's fine. Before AI you still might have pay for such picture, but whatever.
But what if you use it commercially, for an ad campaign, in a game, and so on ?
It's similar issue as with ChatGPT. ChatGPT can't source what it's claiming, and you can never known if it's true or not. Image models can't source either .. and you never know how original the pictures are.

Doubt it, that’s not how they generate new images. Plus surely we’d have seen some examples of this given the literally millions of images that have been produced with generative AI if for some reason it could happen.

[Darat]

Originally Posted by Skeptic Ginger

In my mind that's like saying if an editor edits my novel (beyond punctuation and spelling) it becomes original and unique work.

I am purposefully using text as an analogy because we have more clear guidelines most of the time what is and isn't copyright infringement and what is original work.

The analogy is valid even if the problems of copyrighting styles are going to be immense.

Don’t think the analogy works, let me and try and explain why by expanding the analogy.

In my example the generative AI has created a couple of chapters that match my writing style, vocabulary, grammar (misuse) and plot. Those chapters never existed before, it is a new and unique work and no one can tell the difference between mine and the AI’s chapters. I’m sure that if I presented my second image sorry second chapter no one would be able to tell it was created by a generative AI trained on many copyrighted novels as it would be using my characters, my style, my plot. The AI is not cutting and pasting sentences it’s scanned during its training, it’s using the “knowledge” it’s gained to figure out what the new chapter should be.