ฝัง
- เผยแพร่เมื่อ 25 ธ.ค. 2022
- Today, I'm using OpenAI's ChatGPT to generate music based upon Beethoven's 3rd Symphony. Providing commentary and my own example for reference.
I used MuseScore 4 to render the music: musescore.org/en - วิทยาศาสตร์และเทคโนโลยี
![[Vowel]물고기는 물에서 살아야 해🐟🤣Fish have to live in the water #funny](http://i.ytimg.com/vi/3G9MH063Cx4/mqdefault.jpg)
I'd really be curious about if it could make a 9 voices counterpoint as there are so many things to look for. It's not a thing a human can do without spending an unreasonable amount of time and thinking and it would test if chat GPT can actually form something from texts that may mention rules of counterpoint.
If not, 7 voices.
Thanks for taking the time to explore this a little. I been curious:P
I think for now AI music generation (even from a network specifically designed for it, rather than a general-purpose chatbot) is still at the point where it can suggest lots of interesting ideas, but a human must be there to separate the good from the rest.
In building ChatGPT, they built a system to translate what you're asking it into examples so that it would give you what you want.
I would imagine that no one who was helping build that translation system tried to ensure that it could produce good music outputs.
It may also just plain not understand music if it hasn't been given enough examples of the entire chord progressions for ... many songs... I'm not sure how easy those are to find for free on the net.
At last you've showed us what you're doing in the shadows, magic man
I think orchestral and "serious" music will be safe for a long time, because any AI movement in the direction of music is always going to be geared for pop music, since that's where the biggest market is, and what will have the greatest draw. But thanks for trying this out, at least you proved that ChatGPT has its limits... 😀
Have you heard the AI generated completion to beethovens 10th symphony?
I think that it´s mainly "serious" music that will be replaced by AI in near future. Everything sample-based with a lot of sound-manipulation in the production stage (e.g. hybrid trailer music, electronic dance music or even rather experimental "pop"-music like "bad girl from billie eilish) will be hard to replicate.
@@dingdongsilver4783 No, can't say that I have. Would be interesting.
@@maximilianbeiel6812 Well, it will be interesting to see where the future goes. I remember back in the 90s, a couple of people tried to come up with some software that would write music in the style of Debussy. I never head it, just read about it, so people were thinking about this all that time ago.
You can see what classic song I menage to generate with AI in my channel. I don't think it is safe for long...
Thanks you. Nice demonstration of how to use base, and work keeping it in technologic result.
Thank you!
From what I’ve seen, AI generated music lacks the concept of musical form, which is essential for a composition to be comprehensible.
Training an AI on musical form would also be somewhat challenging, since what makes music interesting the ways it deviates from established structures in unpredictable ways.
I don’t see AI generated music replacing composers until better we have more concise definitions for what makes music good or not.
Extremely underrated!
Wow!!! this is amazing how fast is growing
I totally agree with you!
What happens if you give ChatGPT the notes you need harmony for? Maybe it just hasn't the musical data.
This is insane
I'd like more! This is a new niche of music we are exploring. We need a few explorers to lead us the way ;)
Just don't make some superintelligent AI that destroys the world in D minor please
It will be C flat major, just you watch... 😀
nice!
there is riffusion, an actual music AI software.
it's not producing the same output anymore. It says "I apologize, but as an AI language model I am not able to play or generate music, and I am not able to access the audio files from Beethoven's 3rd Symphony. However, I can tell you that Beethoven's 3rd Symphony is also known as the "Eroica" ...
Seems that the feature has been turned off.
That is interesting to know.
ears bleeding.. id tried the same thing with the same outcome, even tried a simplified prompt, clarifying "Accento" c major, it swapped the third note to an "A", and sounded like nails on a chalkboard.
Damn, I got stuck making music, it's so fucking beautiful with staccato strings.
It may have done better if you had assigned a role at the beginning of the prompt. For example, "you are a highly trained music composer" or even, "you are the famous composer Ludwig van Beethoven".
how did you get musecore to look so good?
I just messed around with the appearance settings. You can change everything from color of the score to fonts.
Whoa, how did you add the notes so fast?
I've been using Musescore for years, I'm rather accustomed to the shortcuts and workflow.
gpt 4 try plz
May be ask him again the same music and give him more specific clue 🤗 style like alot of musician 👍
Thanks for the feed back!
It doesnt generate anything, AI model picks starting notes and caculate probabilities of next notes based on given prompt.
It's actually just tensors and weights, yep. But still I think it's pretty cool.
i knew it! philip glass is ai
How could I have not known!
Writing music does not require intelligence. You just need talent
we can keep our jobs
Should say in the style of
Good point good point.
5:35
cool video, but i think ChatGPT is not the best to make ai music ;-)
What would you suggest?
@@musicallyroy humans…🙄
Not knowing how to ask a proper prompt to AI isnt the AIs fault. If you cant ask the AI the correct question, the AI wont give you your wanted or expected answer... - please try a similar attempt, by giving the notes and not the name of the symphony.
You can dress up and dance around the bedroom. Doesn't make you Madonna. Never will.
it's a language model. So, this doesn't even make any sense. (Even if it did make sense, ChatGPT has no meaning behind it. It's just a parrot trying to mimic what it has read.)
People, dont make the mistake to install this version.. i did and it has a lot of bugs and I mean A LOT!, it´s basically useles..
And yes, my computer meets all the requirements.
As soon as I started using it, the audio is heard at half the normal speed, with echoes and a crackling sound. It doesn't detect me except 2 external vsts, it doesn't read articulations, diamics etc..
Worst of all, now that I wanted to go back to the previous version, it doesn't read the scores and tells me they were made in a newer version...
I lost MONTHS of work... . time to go back to the professionals. NEVER AGAIN MUSESCORE
Have you tried increasing the buffer in IO settings?
@@theelephantseal3142 ^ this should work
Please look into preference guides for musescore 4, it's not perfect but it has a lot of improvements.
I hope this gets fixed!