Microsoft’s AI boss thinks it’s perfectly OK to steal content if it’s on the open web https://www.theverge.com/2024/6/28/24188391/microsoft-ai-suleyman-social-contract-freeware
Microsoft’s AI boss thinks it’s perfectly okay to steal content if it’s on the open web

Mustafa Suleyman, CEO of Microsoft AI, has a curious idea about how copyright works. He incorrectly claimed the moment you publish anything on the open web, it becomes “freeware” that anyone can freely copy and use.

The Verge
@verge did you invent all the words you just used, or did you "steal" them from authors you read in the past?

@verge reward for dumbass of the year:

🏆 (make in plastic)

@verge I'll Say it, loud and clear:

"FUCK THIS GUY AND MICROSOFT!".

Jeez!

@verge AI sure makes a lot of smart people say stupid things

@verge wow i agree actually

that's why i stole every microsoft product i have ever used

@verge That's not the best wording by the Microsoft's employee. But from the text of the article I've got an impression that AI somehow contains copies of your articles. Some program "scrapes a particular website" and copies its content into an AI. That's not how it works. There's no algorithm that makes collages out of copied sentences. If there's no copy, what are you trying to apply the copyright law to? That is the difference between the AI before and the AI emerged a couple of years ago.
@spyke @verge LLMs can reproduce copyrighted works, verbatim. To say “there is no copy” is just sophistry.
@alannorbauer @verge Can they? Or they can reproduce a couple of quotes from the entire article while the rest is its own explanation? Same as humans. Stable Diffusion is like 6 GB in size, but trained on *terabytes* of images. Where did the remaining hundreds of gigabytes go? Clicking "compress into a zip file" doesn't produce such miracles.

@spyke @verge Yes, they can reproduce copyrighted works, verbatim.

https://en.wikipedia.org/wiki/Wikipedia:Large_language_models_and_copyright

See section “Is LLM output capable of violating copyright law?”

Wikipedia:Large language models and copyright - Wikipedia

@alannorbauer @verge If you open your own link, it states "snippets" and "close paraphrases". You can't get the entire thing it saw out of AI. It doesn't exist there physically. Your "reproduce" is vague. Is a Samsung Galaxy S a copy of an iPhone? Is a Verge article a copy of another article if they look similar? The law says no. And AI never copied the text to "change" it to look different. There is no miracle that compresses the entire internet into 10GB. You can't change physics.
@spyke @verge You’re not reading or quoting from the section I referenced which has chatgpt directly quote the verbatim lyrics of a copyrighted song, which is exactly what you’re saying it can’t do. Additionally, you are asserting that it can’t do that because of the laws of physics, when all you have to do is ask chatgpt and you’d see it absolutely can reproduce works verbatim. You’re not letting reality contradict you for some reason. Will this screenshot help?
@alannorbauer @verge Now you can finally explain how this happens if the model is only a couple of gigabytes in size. Another one not that successful:
Listen to the AI-Generated Ripoff Songs That Got Udio and Suno Sued

The record industry has filed a list of thousands of songs it believes have been scraped without permission, and has recreated versions of famous songs using Udio and Suno.

404 Media
@verge yeahhhh well listen... This doesn't work like that..
@verge then let’s steal windows and Xbox games. It’s on the web !
@verge "Boo this man! BOOOOOOOOOOO!"

@verge

This man has an address where we could protest.

He deserves a pie in the face every time he appears on stage.

Every time he tries to eat in a restaurant, someone should take food directly off his plate and eat it in front of him. "Oh, you don't mind if I ingest something of yours, do you, motherfucker?"

The internet is not dying. It's being killed by people with names and addresses.

(Yeah, also, we need really fucking brutal regulation of these bastards, mostly making their whole business model illegal, with guillotine punishments.)