No sense of humour…DM vs Internet

I was very intrigued now what is still possible with udio now after all that...
And of course i was very curious about other aspects of this, so i tried something...

The name of the artist was nowhere mentioned in the process, but it seems if you point it hard enough into a certain direction it still comes up with an exact copy of Artists that fit on that


I guess it cannot get closer than this, and i think this is a flaw in the software.
I did not provide udio with anything else than a prompt.

Edit: or, after thinking about it, it is missing training data. depending what data they use, the further back in time you go ( and especially if you point to an exact year there ) the less data you have in total.
So pointing udio with your prompt to a special year, you get very close to the provided training material.
I have to try this now.
 
Zuletzt bearbeitet:
But they invite all users to report such content:
  • a statement that the information above is accurate, and under penalty of perjury, that you are the copyright owner or the authorized person to act on behalf of the copyright owner; and
  • the physical or electronic signature of a person authorized to act on behalf of the owner of the copyright or of an exclusive right that is allegedly infringed.
But from the last two phrases, I get the clear impression that it's not "all users" that are invited to report, but the copyright owner themselves or their authorized representative? In this case DM or their staff.

Just a question, can you still hear this somewhere? ☺️
Has anyone downloaded this, can someone send it to me or upload it to a cloud? Thx!
Please PM
So I got an answer from the company that hosts my website and they strongly recommended not uploading the videos there, but instead using services meant for such things. So I will follow their advice and won't be uploading them. And probably a good idea not to share the files in general. Sorry.
 
I was very intrigued now what is still possible with udio now after all that...
And of course i was very curious about other aspects of this, so i tried something...
I'm not very civilized with older music. Jimi Hendrix?

With the first DM song, I wrote Depeche Mode as the prompt.

HOWEVER!!! The only prompt I used for all the following "DM" songs was the one that Udio originally replaced Depeche Mode with, which was:

Male vocalist, Electronic, Synthpop, Rock, Alternative rock, Dark, Rhythmic, Melodic, Atmospheric, Nocturnal, Alternative dance, Mechanical, Passionate, Cold, Longing, Sombre

People in the comments said they were getting DM songs with that same prompt as well (as I did share it in the description).
 
what AI could/can do astonishingly well is very minimal (instruments) vocal based music, no matter folk, singer songwriter etc..
with high expression level.

the prompt words: thanks for that!
my guess here is - the system tagged things and has not so many training material that had this kind of signature. let's assume they had the chance of training "all music" might have thrown in more and even more tags.

so if you check that with simpler systems like "people who bought this might love that" seem to be in a similar frame of tags or tag clouds.
really interesting what kind of "intelligence" might be behind genres and tags in general.
 
I'm not very civilized with older music. Jimi Hendrix?

With the first DM song, I wrote Depeche Mode as the prompt.

HOWEVER!!! The only prompt I used for all the following "DM" songs was the one that Udio originally replaced Depeche Mode with, which was:

Male vocalist, Electronic, Synthpop, Rock, Alternative rock, Dark, Rhythmic, Melodic, Atmospheric, Nocturnal, Alternative dance, Mechanical, Passionate, Cold, Longing, Sombre

People in the comments said they were getting DM songs with that same prompt as well (as I did share it in the description).
You shouldn't expect too much from the current version of this type of AI. There is no trace of intelligence and no great variety of results. I wouldn't be surprised if several people with the same or even similar prompts were given completely identical results.
 
my guess here is - the system tagged things and has not so many training material that had this kind of signature.

It really looks like that. to get anything else than Genesis out of it if you point it to 1975 british Prog Rock is very hard.

I am trying different things now since a few hours and it clearly points to a lack of training material. Like they only took one or two of the most well known and successful snippets from every style and year.
And if you point especially to a certain year and style, you only get that.

when you ask for british progressive rock 1970s you get either genesis ( 8 out of 10 ) or early Pink floyd, and it is very hard to get something else out of it. Everything sounds like Peter Gabriel or Syd Barrett.
Only if you especially point to 1978 and later it changes slowly.
 
Zuletzt bearbeitet:
Ich glaube da passiert gerade was ich vor Monaten schon mal in einem ähnlichen Thread prophezeiht habe...
DAS ENDE IST NAH

nein, quatsch.

Text- und Bildmaterial gibt es im Internet genug, und die zu verarbeitende Datenmenge ist mit Sicherheit um einen signifikanten Faktor kleiner als bei Musik oder gar Videos.
Ich dachte eigentlich deshalb wird es noch ein weilchen dauern bis brauchbare Modelle auf dem Markt sind...
Aber wie es aussieht haben die einfach auf eine sehr begenzte Datenmenge trainiert, und es ist ziemlich klar was das Trainingsmaterial war.

Sind Udio jetzt vielleicht mit einem unfertigen Modell vorgeprescht weil Suno mit einem nutzbaren Modell rauskam und die den Hype mitnehmen wollten? Das sind ja offensichtlich keine Big Player wie Meta, Microsoft oder Google.
 
They are obviously not big players like Meta, Microsoft or Google.

They're not a group of geeks in a basement either though. According to this Rolling Stone article:

Udio’s product came together remarkably quickly after its founding last December by four former employees of Google’s AI-research wing, DeepMind — David Ding, Conor Durkan, Charlie Nash, Yaroslav Ganin, and Andrew Sanchez — along with Andrew Sanchez. They’re backed by a range of tech heavyweights, including a16z (a.k.a. Andreesen Horowitz) and Instagram co-founder and CTO Mike Krieger. “We were very well supported from the day we took investment,” says Sanchez. “So the technical co-founders were sort of able to hit the ground running because we could get that all going pretty quickly.”

There are some notable music names on Udio’s list of early investors as well, including Common, producer Tay Keith, industry vet Steve Stoute’s United Masters, and Will.i.am. In a press release, Will.i.am, who’s long been an evangelist for AI’s musical possibilities, was effusive about the company’s product: “This is a brand-new Renaissance, and Udio is the tool for this era’s creativity,” said the artist, who was consulted during the development of the product. “With Udio you are able to pull songs into existence via AI and your imagination.”
 
that's for sure - but they knew how to do it. let others die for you. and when I rush, I rush for you ;-)

well - what I thought is, they only got the Tags rolled out on the more known bands rather than anything underground. that's why synthpop etc triggers davedancing and I'd assume the same for other genres. Editors, Depeche Mode and maybe Hurts are in - others may be there as well - since the new try sounded more like those WGT-synthpop-dark-bands that never got even the idea of what is and was so cool with DM or NEP or …

So - the AI model is about to be hard to prime right.
but in our case it was cool - since I'd really like to check that AI how a generic DM song sounds like - you did.
so - it is sort of our solution for 2024 humanity.

but we must wait.



btw - I tried something "likely" by writing an article about a music style that does not exist and recombined it in a special way.
so now if you ask GPT it throws something at me that I sort of vaccined it with.
enough tags from other pages might re-combine the tags and that might chance how the result could be like?

normally you do not have the chance, here you need to create a band and tags and genre overnight and inject that stuff to the training material and into their engine.
it also shows, that the results are sort of ... hmm - mainstreamed.
that's what I'd like to point out - means - it is sort of not fair - but grabbing stuff is sort of not fair as well since they take the way we post. if Sequencer.de was the source it will and would have responded differently, same here with music.

yes that is the geek inside of me - I bought my first computer because I wanted to unterstand it..
 


Neueste Beiträge

News

Zurück
Oben