Artifical Intelligence and Sharing Knowledge
By now most people in the music and creative industries have at least a vague idea about what AI is doing, and understand that apps like Suno only work because they scraped nearly all artistic output of the past decades…
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
By now most people in the music and creative industries have at least a vague idea about what AI is doing, and understand that apps like Suno only work because they scraped nearly all artistic output of the past decades, and that every result they produce depends on human creations made before it. The imbalance is obvious and glaring: some tech folks getting obscenely rich based on work done by others who get nothing or peanuts. Despicable, if you ask me.
I have no doubt that artists can use these tools, and find ways to make the results personal, meaningful, and radically new. I will not engage in any discussion about this being cheating, unethical, etc. My perspective here also does not fit in a single paragraph.
Instead I'd like to look at another aspect of AI-based creation, one that affects me and my work more directly. As someone working with code, I have seen my working process go through a historic and radical change, one that is not nearly at an end, but certainly way beyond the point of no return. Coding without AI assistance feels like sitting in a car with a steam engine.
For those who are not coding, here comes a brief explanation: Until maybe one year ago, coding meant learning a programming language (or multiple, since a bigger application these days might contain parts written in different ones) and spending a lot of time formalising any task, breaking it down into individual steps and writing a piece of code to do each step.
Imagine a simple audio plugin that does a chorus effect. You need code to draw all kinds of user interface objects on the screen; you need code that can capture the user's interaction with that interface object. You need code that does the actual audio processing. For that part, you might need to read up on the principles of a chorus; you need to experiment with various possible ways of doing it, tweaking and tuning it until it does what you want it to do. Then you need to test it, making sure it does not crash or produce unwanted things at specific settings.
All that takes time, skill and experience. Creating something like a good-sounding reverb is (was!?!) a magic art-science, and the ones who built it were heroes of that specific community and beyond.
As of today, I can simply tell Claude.ai to write me an effect or synthesis plugin. It takes perhaps 30 minutes, and I do not need to have any coding skills for it. Claude can do this because it has scraped the internet for all available papers, code snippets, descriptions, etc. It can also easily reverse-engineer existing products, which means it can derive the underlying code from the results of using a product. Humans could do this too, but it is extremely tedious and requires a lot of expert skills and advanced technologies. Until a few months ago...
Those who follow me a bit more closely might know that I am very open when it comes to sharing knowledge. I have been a professor for sound art at the Berlin University of the Arts; I have been teaching at CCRMA/Stanford University; and I happily answer emails about technical questions all the time.
Since 2011, I have been developing my own laser control software. The latest iteration, the FIRST LIGHT ENGINE, is the result of the experiences with the previous ones, and I put in three or more years of work. I am very, very proud of it. It is the best piece of software I have ever written. It allows me to do things that cannot be done in TouchDesigner or MadMapper or any other commercial software. It's like an intricate complex clockwork, ticking away smoothly and with high precision.
The concert experience is the tip of the iceberg. The software is the rest. My initial impulse is to share that rest below the surface, creating a dedicated page for it, where I explain in detail how it works, what it does, how it is constructed and so on. Similar to what I did for CBM 8032 AV with this page here.
Humans can read this. They can get an idea about the effort behind the scenes. And some might use it as a starting point for their own creations — which is not just fine with me; I actively encourage it. This is how we learn! This is how I learned. I don't mind if anyone takes my ideas and incorporates them into their own tools. Copying an idea and integrating it into another framework is hard work, and it is never flawless. The derivations drive the evolution, and that's great.
I am not going to create a dedicated page for the FIRST LIGHT ENGINE. Not because I am afraid of humans. I am afraid of machines. I don't want all my expertise and experience to become material for AI. I do not want to enable simple reverse engineering of a decade of work. Every description of functionality and concept of my engine is potentially a data point.
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。