"Generative AI is the greatest risk to the human creative class that has ever existed"

[oboemaroni]

Something I've just noticed is multiple new accounts on youtube (most with names that make them sound like labels/studios) posting endless AI novelty songs like the one below. I'm guessing this could be the AI companies themselves putting things out in the wild to see what sticks... Either that or someone with a scatalogical sense of humour and a lot of time on their hands ¯\_ (ツ)_/¯

 

[newbreednet]

see what sticks...

Well played

 

[FluffyTailedRodent]

ChatGPT now has a ‘memory’​

ChatGPT now has a memory of its own, creator OpenAI has announced.
It means that the system will now be able to learn and remember details about its users.

If someone asks for a poem about their pets, for instance, the system may remember that those pets are two cats. It will also remember more stylistic points, such as the fact that someone likes their answers being presented in bullet points, or what computer programming languages they use.

ChatGPT now has a ‘memory’

New update marks major step in AI abilities, and should allow it to be more personalised

www.independent.co.uk

It always had somewhat of a memory when having a conversation tho - I mean that's why it's so easy to crash it once it starts lying about things.

To be honest I haven't even used it in ages because the quality of output has become so bad and it lies and makes up stuff (or hallucinates) way more than it provides legitimate answers that it's basically useless at this point except for stupid things - "Write me lyrics about squirrels dancing in a tree". And even then the output is so infantile that it's basically useless as well, "Let's hold hands and dance with glee that our tails are so furry" and stuff like that. Honestly it makes up stuff and spits out misinformation so much that at this point I swear it's being trained solely on wikipedia. That can be the only explanation why so much of what it spits out is just incorrect.

I wonder if it'll remember how many times I crashed it? Lol maybe I should worry...

 

[Alchemedia]

Studios Warn Of First Amendment Implications For SAG-AFTRA-Backed AI Bill To Require Consent For Digital Voice And Likeness

The draft legislation would apply to living and deceased individuals.

deadline.com

 

[Ed]

Something I've just noticed is multiple new accounts on youtube (most with names that make them sound like labels/studios) posting endless AI novelty songs like the one below. I'm guessing this could be the AI companies themselves putting things out in the wild to see what sticks... Either that or someone with a scatalogical sense of humour and a lot of time on their hands ¯\_ (ツ)_/¯

This isn't at all a surprise if you've seen what people have been doing with AI Images for the past few years. Why wouldn't we think people would make silly songs or silly images. People have made silly songs in these AI threads and they're composers, and some are concerned about AI and they're still making silly songs and other tracks with Udio.

I hope no one's denial is such that that just because you can make silly songs with it means that's all it's good for.

 

[Ed]

It always had somewhat of a memory when having a conversation tho - I mean that's why it's so easy to crash it once it starts lying about things.

To be honest I haven't even used it in ages because the quality of output has become so bad and it lies and makes up stuff (or hallucinates) way more than it provides legitimate answers that it's basically useless at this point except for stupid things - "Write me lyrics about squirrels dancing in a tree". And even then the output is so infantile that it's basically useless as well, "Let's hold hands and dance with glee that our tails are so furry" and stuff like that. Honestly it makes up stuff and spits out misinformation so much that at this point I swear it's being trained solely on wikipedia. That can be the only explanation why so much of what it spits out is just incorrect.

I wonder if it'll remember how many times I crashed it? Lol maybe I should worry...

Are you talking about GPT3.5 or GPT-4?

GPT-4 is substantially better than GPT03.5. Even Microsoft Copilot (previously "Bing Chat") has a GPT-4 model but this is so different to ChatGPT GPT-4 that I really can't consider it the same. I don't know what Microsoft did to it, but it's psychotic.

 

[Alchemedia]

Urgent Action Needed to Stop AI Developers From ‘Destroying’ Artists’ Careers, Says U.K. Parliament Committee

A U.K. parliamentary group calls for the passage of strong AI laws to protect artists from the nascent technology.

www.billboard.com

 

[Ed]

Urgent Action Needed to Stop AI Developers From ‘Destroying’ Artists’ Careers, Says U.K. Parliament Committee

A U.K. parliamentary group calls for the passage of strong AI laws to protect artists from the nascent technology.

www.billboard.com

I hope people realize this would only slow it down until Open Source models get released right?

When a new Stable Diffusion model is released people understand that the true power isn't in the released version, it's in what you can do with it once you start running it with custom trained models.

It would also only apply to UK companies.

This will also undoubtably fail, because we'd have to assume Open AI, Google and Facebook would have to throw out all their models. Does anyone really think this would happen?

What about Getty Images, they sued Stable Diffusion, but then trained an AI on their whole catalogue. Does anyone think their content creators were asked if they want to opt in? How much do you think they got paid, or if they got paid? How much money would you need to be paid if they wanted to buy out your whole industry?

 

[smellypants]

I hope people realize this would only slow it down until Open Source models get released right?

When a new Stable Diffusion model is released people understand that the true power isn't in the released version, it's in what you can do with it once you start running it with custom trained models.

It would also only apply to UK companies.

This will also undoubtably fail, because we'd have to assume Open AI, Google and Facebook would have to throw out all their models. Does anyone really think this would happen?

What about Getty Images, they sued Stable Diffusion, but then trained an AI on their whole catalogue. Does anyone think their content creators were asked if they want to opt in? How much do you think they got paid, or if they got paid? How much money would you need to be paid if they wanted to buy out your whole industry?

It does seem unlikely that all those companies would throw out all their models, but thats exactly what needs to happen imo.

This quote from the article refutes your last point... Again if this bill is passed of course.

"The MU said it was “vital” that any future licensing scheme governing the use of copyright-protected works to train AI systems requires the “clear, explicit consent” of individual creators, as opposed to just the label or publisher rights holders."

And another quote from the article:

"In contrast, the 27-member block European Union, which the United Kingdom officially left in 2020, passed its world-first Artificial Intelligence Act earlier this year. It requires tech companies and AI developers operating in Europe to provide detailed summaries of any copyrighted works, including music, that they have used to train their systems, as well as clear labelling of AI-generated works."

 

[FluffyTailedRodent]

Are you talking about GPT3.5 or GPT-4?

GPT-4 is substantially better than GPT03.5. Even Microsoft Copilot (previously "Bing Chat") has a GPT-4 model but this is so different to ChatGPT GPT-4 that I really can't consider it the same. I don't know what Microsoft did to it, but it's psychotic.

I don't care what microsoft does with AI. They will never top Tay. Tay is the Queen.

 

[AudioLoco]

This will also undoubtably fail, because we'd have to assume Open AI, Google and Facebook would have to throw out all their models. Does anyone really think this would happen?

No, but in a normal industry/sector not plagued by heavy monopoly and too much power, it would have been the immediate thing to do.
You create a product by stealing, you cannot sell the product, make a new one not based on stolen data.
They IDEALLY should throw everything out and start anew.

For some reason tech companies are above the law. And if they get punished they are made to pay fine that are like 0.000000000000001% of what they do in a minute.

 

[AudioLoco]

Something I've just noticed is multiple new accounts on youtube (most with names that make them sound like labels/studios) posting endless AI novelty songs like the one below. I'm guessing this could be the AI companies themselves putting things out in the wild to see what sticks... Either that or someone with a scatalogical sense of humour and a lot of time on their hands ¯\_ (ツ)_/¯

So, this is the only fair use Udio and the likes should have. Ass and farts.
When experimenting with it to test its capability I OBVIOUSLY input only stupid lyrics about farts etc ... So i get it, hahah.....

To create Youtube channels of it and then share it is really next level. The channel is going to make money.
The money is going to be given to the prick that wrote (or did AI did that) the silly lyrics , but the rest, what ideally makes it funny, which is the PERFECT production and PERFECT vocals of amazing talents, and amazing instrumentalists and great studios that have been "training" the model are not going to get a dime though.

Cheap comedy, easy money, AI robbers hit again. There is a taste of the future to come.

Maybe videos using AI music should be flagged for "preventive copyright infringement" everytime they get uploaded.

EDIT: Actually what would be extremely funny, and also a demonstrative action to highlights the new AI legal jungle and deter users, would be:
to download the artwork and songs from these "AI parody" videos and simply create another channel and re-upload them. Or Youtube deletes both or I can do as I wish as there is no copyright involved.

 

[VSTHero]

So, this is the only fair use Udio and the likes should have. Ass and farts.
When experimenting with it to test its capability I OBVIOUSLY input only stupid lyrics about farts etc ... So i get it, hahah.....

To create Youtube channels of it and then share it is really next level. The channel is going to make money.
The money is going to be given to the prick that wrote (or did AI did that) the silly lyrics , but the rest, what ideally makes it funny, which is the PERFECT production and PERFECT vocals of amazing talents, and amazing instrumentalists and great studios that have been "training" the model are not going to get a dime although.

Cheap comedy, easy money, AI robbers hit again. There is a taste of the future to come.

Maybe videos using AI music should be flagged for "preventive copyright infringement" everytime they get uploaded.

EDIT: Actually what would be extremely funny, and also a demonstrative action to highlights the new AI legal jungle and deter users, would be:
to download the artwork and songs from these "AI parody" videos and simply create another channel and re-upload them. Or Youtube deletes both or I can do as I wish as there is no copyright involved.

Those AI novelty songs listers are the first YouTube threads I’ve ever blocked .

 

[AudioLoco]

Those AI novelty songs listers are the first YouTube threads I’ve ever blocked .

It's the first time I've seen them, it was just waiting to happen...

Do you imagine with no AI, deciding to do these songs, writing a period correct melody and hook, dig up all vintage equipment, find musicians who can play this music properly and have the right instruments, record all the instruments, mix really well to make it sound vintage correct, find a singer that can rival Marvin Gaye and other icons.
Then let them sing about farts.

It would take a lot of skill, knowledge, time, space, people (maybe even money) to do one of those tracks.
You would need to be really committed to really go through all this.
Or probably you just couldn't be arsed....
With AI it's all free, no time or work required... Just throw it on Youtube and collect the checks.

Don't just block, report!

 

[Sfpx]

Interesting video with some interesting history

 

[MarcusD]

AI Mario anyone?

When you read through video comments, it’s interesting to note most really don’t care about the negative effects AI will have. Too busy being bamboozled asking for their next fix.

 

[holonology]

The one thing that I've always felt is that regardless of the tool, I think people want to know that there is a person behind creative stuff. To some extent AI might, or might not be able to generate art that creates that connection. But like that Mario Video above, I almost want to know who made it, and why they made it. And also, if AI is taking over output, that seems a bit odd when 90% of video on Youtube is basically commentary. Rather than show a music video, you'll get multiple "reacts to" videos, and "hears for the first time" videos. After the pandemic, I don't remember people saying that after 2 years of reduced human interaction, what they wanted was less of it.

A lot of people say creatives will all end up just pumping queries into the AI tools. And there is some evidence of that up to a point, but within very specific constraints. But to come back to the example, the Mario video above had a thought behind it. And it's that thought, that person and that concept that creates the intrigue. You can 100% use AI to do some of it, but the actual core idea itself comes from the same place. Maybe most people will be in the future, happy with a simulated connection to another person. I'm just not completely sure....

 

[VSTHero]

I can give a great example of

The one thing that I've always felt is that regardless of the tool, I think people want to know that there is a person behind creative stuff. To some extent AI might, or might not be able to generate art that creates that connection. But like that Mario Video above, I almost want to know who made it, and why they made it. And also, if AI is taking over output, that seems a bit odd when 90% of video on Youtube is basically commentary. Rather than show a music video, you'll get multiple "reacts to" videos, and "hears for the first time" videos. After the pandemic, I don't remember people saying that after 2 years of reduced human interaction, what they wanted was less of it.

A lot of people say creatives will all end up just pumping queries into the AI tools. And there is some evidence of that up to a point, but within very specific constraints. But to come back to the example, the Mario video above had a thought behind it. And it's that thought, that person and that concept that creates the intrigue. You can 100% use AI to do some of it, but the actual core idea itself comes from the same place. Maybe most people will be in the future, happy with a simulated connection to another person. I'm just not completely sure....

I can give a great example of this. I had family visiting from overseas this past weekend and spoke primarily French with them (though I speak like a 5 year old in French). But I improved rapidly and certain words have stuck with almost no review because there was so much story, personality, and meaning in the interactions.

I also just discovered that ChatGPT 3.5 now has a voice version, where you can engage in conversation with it. The voices are incredibly well done, you can pick any topic, it speaks remarkable French and it is a great tool; but it's still a world apart from talking with a real person and the experience (while still helpful), is way less impactful - it feels more like work/study than a lived immersion experience.

We are social by nature and our bodies/minds respond differently to real experiences with others. AI is several layers removed from that.

 

[Ed]

AI Mario anyone?

When you read through video comments, it’s interesting to note most really don’t care about the negative effects AI will have. Too busy being bamboozled asking for their next fix.

This is the reason powerful people will never side with “all models need to use cleared copyrighted material/copyright free works”.

It’s just too good.

Open AI would have to throw way their whole model, including their new models. But most people aren’t so viscerally aware that these LLM’s are trained the same way the Image and Audio models are trained.

I don’t know if you’ve seen it yet, but this is Sora. It’s far more advanced than any of the competition, and it’s been months since they finished it. They’re already working on Sora 2. There’s plenty to improve. Like it doesn’t currently “truly” understand the physics, so the “world” doesn’t always respond in a realistic way.

Sora makes current AI video look like Dall-1 or Midjourney Version 3 (or even 2).




If you think the start of the above film “The Hardest Part” looks a little odd, it looks like Paul intentionally left it looking like that for artistic reasons.

It looks practically perfect. And this is just version 1 of Sora. There’s plenty to improve, but if you can’t see where this is going to go very quickly then you don’t have any vision.



Sora was trained on copyrighted works.

They’re never going to let a court rule that they have to junk all of this. Setting a precedent with music applies to everything else as well.

I would be surprised if they don’t have a Music Ai that’s way better than Udio they’re keeping secret for now.

 

[Ned Bouhalassa]

This is the reason powerful people will never side with “all models need to use cleared copyrighted material/copyright free works”.

It’s just too good.

Open AI would have to throw way their whole model, including their new models. But most people aren’t so viscerally aware that these LLM’s are trained the same way the Image and Audio models are trained.

I don’t know if you’ve seen it yet, but this is Sora. It’s far more advanced than any of the competition, and it’s been months since they finished it. They’re already working on Sora 2. There’s plenty to improve. Like it doesn’t currently “truly” understand the physics, so the “world” doesn’t always respond in a realistic way.

Sora makes current AI video look like Dall-1 or Midjourney Version 3 (or even 2).




If you think the start of the above film “The Hardest Part” looks a little odd, it looks like Paul intentionally left it looking like that for artistic reasons.

It looks practically perfect. And this is just version 1 of Sora. There’s plenty to improve, but if you can’t see where this is going to go very quickly then you don’t have any vision.



Sora was trained on copyrighted works.

They’re never going to let a court rule that they have to junk all of this. Setting a precedent with music applies to everything else as well.

I would be surprised if they don’t have a Music Ai that’s way better than Udio they’re keeping secret for now.

I think I read that it’s called Jukebox.