gpt-3 was able to generate guitar tabs, albeit sloppily, when I prompted it with an example. i would be interested in seeing a talented (or even a not so talented) individual attempt to collaborate on a song with GPT-3 as the co-composer. assuming it is capable.
im also curious if gpt-3 can generate ascii art. so far i haven’t found it useful to copy and paste off the internet, the font formatting doesn’t seem to transfer well.
just some ideas to kick around.
very interesting read. thank you for sharing. i would have to agree that openai’s text collection methods do seem to need work. sometimes i get off-completions that read like they were ripped directly from a comment section.
I’m also a musician and plan to release songs on a regular basis with GPT-3 as the muse. Staring at a blank page is frustrating as hell.
I’m mostly planning to use it as lyric generator for ideas on second line or a second verse. For instrument I’m trying chord progression generation rather than the tabs as I guess it would give better results.
how has that worked for you? anything interesting? i thought about doing chord progressions, too, but I couldn’t imagine it doing anything different from conventional chord progression generators.
Nothing too interesting yet. Most songs already use the I–V–vi–IV chord progression, so maybe it doesn’t matter that much 
But it can be helpful to have choices on where to go next, and it might work better than generators with coded music theory.
maybe if you cranked the frequency and presence penalties up to the max it’ll be forced to get more creative.