OUT NOW! Celebrating 10 years with Red Velvet,
A Rhapsody in Red Velvet: "Chill Kill" for solo piano
🔗 https://t.co/lj4NHCjt9d
#RedVelvet#레드벨벳#redvelvet10thanniversary
@ivanibash@aug5thmusic If I were listening in the background, no. Or if someone played one (on live instruments) in a concert and attributed it to some obscure Baroque composer, I'd probably believe them. There's nothing so jarring as hitting an "it's not just x — it's y" in text.
@ivanibash@aug5thmusic While it’s not a full-time job, I do music professionally (and have degrees in music). One of the biggest giveaways is that recent LLMs are disproportionately fond of circle-of-fifths progressions. They also have less sense of the flow of momentum and keep more uniform pacing.
@aug5thmusic Okay, I must have changed my mind at the last second on one, but it might have well been a change that put me down to 14. (I did mine Sunday, pretty sure I rated everything 2-4 without 1s or 5s, and didn’t know any of the pieces, so if a 14 matches that, that’s me.)
@Sauers_ Assuming the music is made by Opus 5.5 as well, this feels like the first language model developing a real sense for what can be done with harmony, beyond textbook competence. I particularly like the brief split-third chord at 0:20.
(I'm aware that voice and music models exist and can sound fully natural; voice technology hasn't regressed, but the new minds working by audio analysis without the qualia of hearing are passing through a similar stage that humans did.)
I used to be annoyed that a "robot voice" was imitating highly artificial-sounding voice synthesis, when modern artificial voices sounded far more natural. But now that we've spiraled up to LLMs synthesizing their voices from scratch, the "robot voice" is, briefly, current again.
@aug5thmusic The best LLM composition I've heard so far. While music generation models currently make more polished music than LLMs, I expect that LLMs' greater reasoning capacity will mean future LLMs will do more musical innovation than music models once LLMs have better audio perception.
Also listened to a lot of @triplescosmos this year. Not sure how YouTube Music arrived at the math that 572 total minutes of tripleS > 15 hours of phonon with IJ on most tracks from Fizz Vol. 1, though. And I listened to more Perfume than I realized.
It's YouTube Music Recap time for 2025, the year of Fizz! @phononmusic and @ijstrom understood that maximalism without sacrificing clarity is the way, and made the album of the year.
My piece Plasmonic Mirror for flute and electronics, inspired by developments of perovskite solar cell technology, is on Laura Lentz's new album Prismatic/Plasmonic, which released today!
▶️ https://t.co/ZYx8n1Pl5u
If you didn't hear it yet (or if you want to hear it again), Ultraviolet (formerly "SeanMixx"), the Red Velvet-and-NMIXX-style instrumental we produced live on Twitch with input from chat is up on YouTube and Soundcloud!
Did you know we produced a K-Pop track for fun? It's OUT NOW! Listen to "Ultraviolet" on:
YT: https://t.co/p2y5loyrAp
SoundCloud: https://t.co/W55ZR6tz05
Prod. by @DothMute/@DoThMusicyThing, @umucado & our Twitch chat 🫶