So how are things like Udio working if not a type of LLM?
Discussion
I don't know how it generates the instrumental part, but it's not an LLM.
Fair enough, I tend to say LLM as it’s not a form of intelligence, it’s just compiling everyone else’s work.
It doesn't matter what you call it or what you think about it if it helps you do a task faster or with less effort.
I use a couple of machine learning tools marketed as AI for that reason but that's different to making music by typing a handful of commands.
So what do think about voice conversion? It won't stop till der Fuehrer covers every song in existence.
Like the LLM machines, they are layers of transformer networks sandwiched between layers of attention networks.
Unlike LLMs, the music machines are trained on music not language.
The phrase for describing both/all is Generative Pre-Trained Transformer, GPT, even though "GPT" from OpenAI itself is only an LLM, and there are many other types of pre-trained transformers not trained on language.
Computer jargon is always stupid.
Yeah I'd thought that whilst music is essentially a language of combined sine waves, it's something under the banner of an LLM but I'm too much away from the sources of input on such things, appreciated the explanation!