[sdiy] ML / AI for music & synthesis (not coding!)

[Mike Bryant]

In the 60s there were statements against early sampling (I still hate Mellotrons but that'sjust me) and then we had "no-synths" Queen. Later on it was digital sequencer in the hate headlights.

Yet now all are an accepted part of creating music.  LLMs will be the same in due course.

On 25 Apr 2026 12:29, Florian Anwander <fanwander at mnet-online.de> wrote:
Am 24.04.26 um 20:45 schrieb Eric Brombaugh via Synth-diy:
A more interesting topic to me is the use of machine learning and AI methods for realtime synthesis / effects and music creation.
Hmm, there we are again, where Roman wrote "I really like to know exactly how the thing I'm making is doing what it's doing in every detail." Also in synthesis for me it is the most satisfying part, that I know what I am doing. I don't want to use, what comes from the blackbox - even if it sounds "cool". I want to have it on my own.



Am 25.04.26 um 12:18 schrieb Sean Ellis via Synth-diy:
There is nothing in the current LLMs that help me.

Here are a few more observations on what LLMs are capable of and on what not. I work (or rather, worked) in technical support for enterprise ticketing software that is used, among other things, for ... technical support. We introduced an API for LLMs quite early on (four years ago) and tried to use it for our own tasks as well. Our goal was to create generic HowTo documents based on individual support tickets that had been successfully resolved. Unfortunately, we found that while LLMs can do many things, there is one thing they cannot do: take a specific case and derive the general problem from it. An LLM can derive a solution for an individual problem from generic knowledge it has previously learned. But an LLM will never recognize the generic concept behind a specific case. That requires intelligence, and LLMs are not intelligent.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20260425/5960ca8f/attachment.htm>