Why is music so much easier for AI than code? Guys I made this song on Suno ai and it is literally, to me, more beautiful than any song I have ever heard in my life, ever. But I still feel that I am a better coder than even the best models. (As of gpt 5.4) It seems that music is so hard, why is it ai has already surpassed us there but not in code? Seems that code is easier. https://suno.com/s/ZpeRr0nngDrlef5D |
There is no way this is an honest statement, unless this is the first song you've ever heard. I can only assume you're a bot from the marketing department at Suno.
I note by this that you don't seem to have experience in music? Because in that case, you are not equipped to judge the quality of the song at some levels beyond a listen.
The same way that someone not technical can't judge the code output of the LLM and will just be impressed by the functional output (i.e. is the site working)
Pick a scale and a tuning, a rythme signature, some chords and a progression , some random melodic pattern, a set of instruments that fit the target genre, and it will sound okay.
Now, it’s much harder to make original and musically interesting music.
cc: @dang
I like messing around with Suno, but I know I don't have much musical taste, yet even so when I try to push it to be even a little weird it fails utterly: a call for "one minute" of animal noises is rendered as 90 seconds of electric guitar; asking for every verse to be in a different accent from around the world is rendered as *at best* two, but usually one; asking for one specific British regional accent often comes up American, and when it actually is British it's one of two specific accents that's either "posh person trying to be middle class" or "posh person trying to be working class".
I was mostly happy to turn out a meme version of the recent SoftBank slides as a parodic James Bond theme, but even then, no goose honk noise.