I do want to say just because there is a different way of making music available doesn't mean the old ways are done.

You can still use an analog synth to write a synth line. Or you can use a Serum preset.

You can record to tape. Or you can use a Daw

You can upload a snippet of a song you are working on, extended it with a prompt, get inspired and sample it, re play it, or even get into a flow state based on what was generated.

All of these are suggestions. All of these options might be a path for you to get your creative ideas out. None of them take from the other none are "better" than the others.

Create how you want to with the tools you want to use.

 

Fine, have it your way. Then it's all about AI-generated music and the music industry. The music industry is all about who owns what rights to what and how they can profit from it. The copyrights holders are first and foremost the majors and they will do everything they can to take advantage of this technological development and squeeze every single cent they can out of it for as long as they can. You think it's over? Well, I disagree. Like I said, it hasn't even really started yet.

Sure, but you’re skipping the basics. No one does expect a 1-day-old newborn to deliver a sprint at the Olympic level. First, the baby must learn to crawl.

The application of music generation is essential to teach an AI the essence of music. First of all, you had to learn what music is, for that you had to listen to music. To understand what makes genres like Metal or Hip-Hop, you first had to listen to music of these genres. The more you listen to different genres and the more you explore the music of each individual genre, the better you will be able to distinguish them from each other.

For an AI to execute commands given by people, it must learn to communicate with people.
For an AI to understand spoken words, it must learn to recognize spoken words.
For an AI to make music, it must first learn what music is and what constitutes it.
For an AI to operate a computer like a human, it must first learn how a human operates a computer.

Many people often forget in discussions about artificial intelligence that the components of their own intelligence - that is for example, their knowledge, problem-solving ability, and skills - did not simply fall from the sky, but they acquired these over many years with a lot of independent and guided multimodal training. Even seemingly simplest things like “seeing” are not given for granted and had to be learned first.

A script is a series of predefined instructions that are executed in a predefined sequence based on predefined rules. Trained models of AI algorithms are regularly incorporated into scripts, but one is not the same as the other. So please don't muddy the waters.

I don't think I said anything like there are "many definitions and perspectives on AI". I said there are different definitions of intelligence. And even the most conservative interpretation model of intelligence divides intelligence into fluid intelligence, crystallised intelligence and acquired skills. However, there are many other models of intelligence that divide the concept of intelligence into many more subcategories. And because there is not one correct scientific definition, but many scientific definitions of intelligence, and different people understand intelligence in different ways, it is essential to specify the definition you are referring to before discussing intelligence. Otherwise, any discussion about intelligence revolves around nothing.

You're thinking too narrowly. And I don't mean that as an insult.

First of all, I have to add to your comments on the limit of this opitimisation: as of today.
However, your basic assumption is not correct. Firstly, we don't know what tomorrow will bring, and secondly, there are already plenty of options for optimisation.

On one hand, AI algorithms and models run on GPUs because they can run a wide range of applications and can perform a large number of calculations in parallel, resulting in high throughput and low latency. However, GPUs are not specifically optimized for specific AI applications. Different applications have different requirements. Research is currently underway to develop hardware that is specifically designed and optimized for certain AI applications and requires less power.

On the other hand, the software side is constantly being optimized. In the past, huge amounts of data were required for training, with efficiency hardly playing a role. How training data will be generated and processed in the future is still unclear, but energy and data efficiency are a priority for future models. More data and therefore more power consumption does not automatically lead to better models.

So there is a lot of potential for optimization in these two areas. I think it is therefore very comparable with the example I gave earlier.

 

Last edited: Jul 4, 2024 at 2:22 AM