Wriffs
Back in 2021, I started my “riffs” series, where I would record my ideas either as an audio file or as a video, and then would take the transcripts and use them as a basis for my writing. But there's a difference between writing done as dictation, and writing done as we normally do.
What's interesting now with LLMs is that you can have them analyze your transcripts. They make them very concise–more concise than humans can do because humans tend to ramble and have a less capacity for clear linear thinking.
For this one, I had done a series of voice memos on my phone over a number of days on the idea of instruments being containers for music. I like how AI found patterns within the various audio files, and has changed my thinking in some ways.
So it could be that we're entering a new era where we're expressing our ideas not in written form, but as simply talking and making videos, and the text that it creates is a byproduct.
But I still think that writing is actually better than riffing. For the transcript, I did go through and I tightened it up as a traditional piece of writing. (Not the raw text that the LLM used).
Something similar is happening with AI music that we create either from lyrics that we've created, or are generated by AI, and is decidedly different than creating music manually. With generative AI, I think we're more likely just to go along with whatever it produces because it's good enough. not the best that it could be, although “rambling” is in fact improvisation. Imagine taking the best jazz solos and having AI “tighten them up”--cleaning up, or “quantizing” them. I've noticed that AI music does this, where it takes the natural flow of language and moves the beats around in an unnatural way. We naturally ramble as humans, and that’s perhaps a good thing. The happy medium is the “wriff” something that has the quality of being both spoken and written, as this was: dictated and edited simultaneously. The bad thing is that it becomes too verbose, and AI is good for condensing it.
Comments