Hacker News Reader

Comment by mikkergp | Hacker News Reader

Dall-E for music will be interesting, but I think it's different. In a sense I think we're already there. Not in the sense that AI makes music but that music is so abundant, and there already aren't a ton of jobs in music writing. It's not a trade in the same way that like graphic design is. I mean, maybe Hans Zimmer loses his job but socially it doesn't seem like that big of an impact. Musicians don't tend to make money from streaming, and if you like going to shows to see performances, you're probably not going to watch a server rack on stage (Maybe, who knows what the future will bring!).

View on HN

A "Dall-E for music" will put much of the control into hands of listeners. That is, you will not search for the music that matches your mood, you will ask for it directly, and maybe adjust in near-real time.

A DJ will arrive with a unique set, likely with every track custom-made for a given gig.

Selling any records at all will become very-very hard, except for rare hits with outstanding human vocal performance. In music clips, music will be relegated to the position of a movie soundtrack, if not lower.

I suspect ai music will have the uncanny valley/98% done problem for a while. For 1 I suspect the 'DJ' in your example, being an actual DJ or the artist themselves plays a larger part in how people listen to music, especially when it comes to 'pop music' (it may be less so for electronic/classical/jazz/"artistic" music.

Obviously for anything sort of focus-y, house music downtempo etc. If we're not already there, we'll probably be there soon, though I am curious if a careful listener will eventually notice the uncanny valley problem there. But pop music I'd say has two problems. 1. There's a je ne sais quoi quality that's hard to replicate, and two I imagine the corpus is just not that big. I mean sure, there's a decently large corpus of pop music, but good pop music? how many hip hop billboard charts have there been a thousand, maybe a few thousand. How do you combine Beyonce, Janet Jackson, Mariah Carey and Whitney Houston into a banger that doesn't sound too much like Beyonce', Janet Jackson, Mariah Carey and Whitney Houston without the titular character marketing said music.

I hadn't thought about it until just now, but soundtrack music has been so terrible for the last 15-20 years that it's one area AI might genuinely be much better than what we've got now. Be hard to do worse, anyway. Studios and producers don't want to pay for good music anymore, so maybe they can get so-so AI music for cheap, and at least it'll be better than the crap they're using now.

Well, you might not want to watch a server rack, but what if that server rack were eventually powerful enough to run a light show, splice a video montage, and compose a song in real time, together based off of audience feedback? I've gotta believe that people would show up just for the spectacle.

Reply on news.ycombinator.com