Data Poisoning could be a tool we use to identify AI that has used copyritten material

https://www.youtube.com/watch?v=Z8aLGHmnRyc

Data Poisoning could be a tool we use to identify AI that has used copyritten material, or we use it to mess with AI.

https://www.vice.com/en/article/infinite-ai-homer-simpson-cover-songs-poisoned-soulseek/

https://djmag.com/news/spotify-and-major-labels-sue-shadow-library-over-alleged-scraping-of-86-million-tracks

https://mosis.eecs.utk.edu/harmonycloak.html

https://mosis.eecs.utk.edu/publications/meerza2024harmonycloak.pdf

https://www.eff.org/cyberspace-independence

131 points · 14 comments · view on lemmy.world

14 Comments

cypherpunks@lemmy.ml · 30 pts · 112d (8 replies)

identify AI that has used copyrighted material

but, that is basically all modern "AI".

(the only LLM i've heard of which actually claims that its training corpus is freely licensed is Apertus...)

youcantreadthis@quokk.au · 8 pts · 112d

We callin it Plagarized Information Stochastic Stupidity now the only PISS you've heard of

Hackworth@piefed.ca · 4 pts · 112d (1 reply)

Adobe claims to only train their image generator, Firefly, on images from their stock library.

very_well_lost@lemmy.world · 7 pts · 112d

lol

cloudskater@piefed.blahaj.zone · 3 pts · 110d (1 reply)

Even if it didn't use copyrighted stuff, the concept of "generative" AI is fascist to its very core.

HaraldvonBlauzahn@feddit.org · 1 pts · 110d

the concept of “generative” AI is fascist to its very core.

Can you explain? I might miss the connection.

YourMomsTrashman@lemmy.world · 2 pts · 111d (2 replies)

Traditionally, with machine learning, it is standard practice to mention what datasets and/or pretrains were used, so that the results are transparent and can be replicated. With GPT-2, it was "the common crawl and our own crawled 8 million web pages", and since then I feel it's mostly left out, falling back on (easily manipulated) benchmarks instead 😬

cypherpunks@lemmy.ml · 2 pts · 111d (1 reply)

Yep. But just providing a list of millions of URLs and saying "we trained on this" as some models in the past have done also didn't make it possible to replicate; by the time anyone re-fetches them all, many of the URLs will inevitably have changed or disappeared.

YourMomsTrashman@lemmy.world · 1 pts · 111d

That's exactly why projects like the common crawl exist though !

very_well_lost@lemmy.world · 14 pts · 112d (1 reply)

People have actually been doing this to catch plagiarism for centuries, long before LLMs were a thing.

See trap streets for one of the better known examples.

LodeMike@lemmy.today · 3 pts · 112d

I learned about that from Doctor Who!

sidefaceturdtalker@leminal.space · 4 pts · 112d (1 reply)

One way to push back forsure but we need to refresh the tree of liberty asap

ZDL@lazysoci.al · 1 pts · 111d

Interestingly, literally zero of the people I've seen who word things this way ever seem to volunteer to be the ones doing the watering. Are you going to break the losing streak or are you going to continue confirming my belief that it's only chicken hawks who say this?

ExtremeDullard@piefed.social · 2 pts · 112d

This idea is as old as books.