caption
a screenshot of the text:
Tech companies argued in comments on the website that the way their models ingested creative content was innovative and legal. The venture capital firm Andreessen Horowitz, which has several investments in A.I. start-ups, warned in its comments that any slowdown for A.I. companies in consuming content “would upset at least a decade’s worth of investment-backed expectations that were premised on the current understanding of the scope of copyright protection in this country.”
underneath the screenshot is the “Oh no! Anyway” meme, featuring two pictures of Jeremy Clarkson saying “Oh no!” and “Anyway”
screenshot (copied from this mastodon post) is of a paragraph of the NYT article “The Sleepy Copyright Office in the Middle of a High-Stakes Clash Over A.I.”
This is simply not stealing. Viewing content has never ever ever been stealing.
There is no view right.
They are downloading the data so thei LLM can “view” it. How is that different than downloading movies to view them?
They’re not downloading anything tho. That’s the point. At no point are they posessing the content that the AI is viewing.
This is LESS intrusive than a Google web scraper. No one trying to sue Google for copyright for Google searches.
What? Of course they are downloading, the content still has to reach their networks and computers.
Go look up how ai works. There is no download lol. It’s the exact same principal as web scrapers which have been around for literally decades.
Tech illiterate guy here. All these Ml models require training data, right? So all these AI companies that develop new ML based chat/video/image apps require data. So where exactly do they? It can’t be that their entire dataset is licensed, isn’t it?
If so, are there any firms that are using these orgs for data theft? How to know if the model has been trained on your data? Sorry if this is not the right place to ask.
You know how you look at a pic on the internet and don’t pay? The AI is basically doing the same thing only it’s collecting the effect of the data points ( like pixels in a picture) more accurately. The input no matter what it is only moves a set of weights. That’s all. It does not copy anything it is trained on.
Yes it can reproduce with some level of accuracy any work just like a painter or musician could replay a piece they see or hear.
Again, this is not theft any more than u hearing a Song or viewing a selfie.
I make the exact argument all the time and it gets ignored. I think people fundamentally don’t understand the tech and can’t conceptualize that AI models train the same way we get ideas and schemas from our own observations.
People even deny that AI can “learn” but that they just copy and manipulate data…
Isn’t that the entire point of creativity. though? What separates an artist from a bad painter is the positioning of pixels on a 2-Dimensional plane? If the model collects the positions of pixels together with the pixel RGB (color? Don’t know the technical term for it), then the model is effectively stealing the “pixel configuration and makeup” of that artist which can be reproduced by the said model anywhere if similar prompts were passed to it?
Focus. We are talking about copyright. Copyright doesn’t cover this at all.
That’s one thing, but I think regurgitating it and claiming it as your own is a completely different thing.
Also, I’m pretty sure the argument is more about the unequal enforcement of the law. Copyright should be either enforced fairly or not at all. If AI is allowed to scrape content and regurgitate it, piracy should also be legal.
Again that’s not what’s happening here
Could say piracy is just running a program that “views” the content, and then regurgitates its own interpretation of it into local data stores.
It’s just not very creative, so it’s usually very close.
You could say that but you’d be wrong.downloading is a bitwise copy. Training isn’t even close to the same thing.