18.08.24
gen ai doesn't know how or when to shut up. We still only have access to toy A.I so I hope your not using hallucinating LLMs for anything serious. You have to manually inspect and fact check LLM output because its not sentient and doesn't understand if it's output is correct. It's a chatbot on steroids, its basically an army of monkeys on typewriters that can output the complete works of shakespeare by accident given enough time.
I've always found silence profound. I remember being in primary school and being asked what my favourite music was. My answer was "silence" and I remember the teacher took a moment to wrap their head around the answer they probably weren't expecting. Silence is beautiful and is the probably the direct opposite of noise or music. Only human beings can appreciate it.
The use of silence is poor with music generators, asking an image generator to make a blank image is difficult. Asking an LLM to return nothing doesn't work which is why people are exploring "structured" output from LLMs. This is inherently problematic because LLMs are good at working with inconsistent input and can only give inconsistent output.
I have been unable to get a music generator to produce anything close to silence except by accident. The prompt is interpreteted as a wildcard and you basically get anything. If you give an LLM or ML generator an ambiguous prompt enough times you will eventually iterate through its training data unless the LLM has guard rails or a restrictive system prompt.
You can't really ask A.I music generators to make compositions that include silence via prompts. They are terrible and understanding these kind of prompts. Try asking a musicGen for something like "a single wood block hit with 5 seconds of silence inbetween". These systems cannot take that kind of instruction yet and will just give you whatever the hell they want (basically a wildcard response). Maybe this level of sophistication will happen in the near future and put human musicians and composers out of business entirely. Or maybe a new vista of human creativity that employs these tools to make a new and wierd form of music might occur which I'm hoping will happen as an alternative. Actually following instructions correctly in a timely mannor and in the context is an incredibly complex thing that only intelligent sentient entities can probably do.