There is no ethical generative AI. Meaning fully permissioned datasets, end-to-end. Not yet scientifically possible. So 100%, everyone who claims this, is lying, usually by omission, and some BS startup isn't going to invent this.
In my open letter, I wouldn't say "ethical" or "environmental" or any of these intersectional things because you're giving space for lies.
People want ethical AI even if it's impossible. So we get aspirationally ethical AI. Meaning, people really want to use generative AI, it makes life so easy, and people also want it to be ethical, because they don't want to make others upset, so they will buy into a memetic story that it is "ethical." Even if that story isn't true.
Aspirationally ethics already got hundreds of millions of dollars in funding. Like look at generative AI in the media industry. Moonvalley - "FULLY LICENSED, COMMERCIAL SAFE" (https://www.moonvalley.com) - and yet, what content was their text encoder trained on? Not "fully licensed," no not at all. Does everything else they make work without a text encoder? No. So... But people really want to believe in this. And it's led by DeepMind people! Adobe has the same problem. Some efforts are extremely well meaning. But everyone claiming expressly licensed / permissioned datasets is telling a lie by omission.
It's not possible to have only permissioned data. Anthropic and OpenAI concede, there's no technology without scraping. Listen, they're telling the truth.
In my open letter, I wouldn't say "ethical" or "environmental" or any of these intersectional things because you're giving space for lies.
People want ethical AI even if it's impossible. So we get aspirationally ethical AI. Meaning, people really want to use generative AI, it makes life so easy, and people also want it to be ethical, because they don't want to make others upset, so they will buy into a memetic story that it is "ethical." Even if that story isn't true.
Aspirationally ethics already got hundreds of millions of dollars in funding. Like look at generative AI in the media industry. Moonvalley - "FULLY LICENSED, COMMERCIAL SAFE" (https://www.moonvalley.com) - and yet, what content was their text encoder trained on? Not "fully licensed," no not at all. Does everything else they make work without a text encoder? No. So... But people really want to believe in this. And it's led by DeepMind people! Adobe has the same problem. Some efforts are extremely well meaning. But everyone claiming expressly licensed / permissioned datasets is telling a lie by omission.
It's not possible to have only permissioned data. Anthropic and OpenAI concede, there's no technology without scraping. Listen, they're telling the truth.