News for hackers. Stuff that matters.
Re:Tencent Hunyuan-Large by ronsor on Tuesday November 05, 2024 @04:24PM (#42055297)
Encyclopedias may be collections of facts, but the writing is generally creative. Phone books are literally just facts. AI models are literally just facts.
Re:Tencent Hunyuan-Large by margalabargala on Tuesday November 05, 2024 @05:37PM (#42055803)
> AI models are literally just facts.

Are they, or are they collections of probabilities? If they are probabilities, and those probabilities change from model to model, that seems like they might be copywritable.

If Google, OpenAI, Facebook, and Anthropic each train a model from scratch on an identical training corpus, they would wind up with four different models that had four differing sets of weights, because they digest and process the same input corpus differently.

That indicates to me that they are not a collection of facts.

Re:Tencent Hunyuan-Large by ronsor on Tuesday November 05, 2024 @06:49PM (#42056238)
The AI training algorithms are deterministic given the same dataset, same model architecture, and same set of hyperparameters. The main reasons the models would not be identical is due to differing random seeds and precision issues. The differences would not be due to any creative decisions.
Re:Tencent Hunyuan-Large by margalabargala on Wednesday November 06, 2024 @12:00PM (#42065303)
Sure, but they don't all use the same algorithm, the same hyperparameters, etc.

At some point, with sufficiently many hyperparameters being chosen, that starts becoming a creative decision. If 5 parameters are available and all are left at the default, then no, that's not creative. If there are ten thousand, and all are individually tweaked to yield what the user wants, is that creative?

Not to mention all of these companies write their own algorithms to do the training which can introduce other small differences.

Re:Tencent Hunyuan-Large by roywiggins on Tuesday November 05, 2024 @04:33PM (#42055369)
What if I train an AI model on exactly one copyrighted work and all it does it spit that work back out?

eg if I upload Marvels_Avengers.mkv.onnx and it reliably reproduces the original (after all, it's just a fact that the first byte of the original file is OxF0, etc)

Re:Tencent Hunyuan-Large by bdowling on Tuesday November 05, 2024 @06:00PM (#42055951)
A work that is “substantially similar” to a copyrighted work infringes that work, under US law, no matter how it was produced. (Note: Some exceptions apply and you have to read a lot of cases to get an idea of what courts find “substantially similar” .)
Re:Tencent Hunyuan-Large by HWR_14 on Tuesday November 05, 2024 @10:48PM (#42057110)
> no matter how it was produced

IIRC, this is wrong. Independent creation is a valid (but almost impossible to prove) defense in US copyright law.

This example is not an independent creation, but your reasoning seems wrong.

Re:Tencent Hunyuan-Large by bdowling on Friday November 08, 2024 @01:13AM (#42084605)
I wrote "some exceptions apply" to try to avoid getting into the weeds, but yes, independent creation is an exception. Other exceptions include out-of-term works, public domain, Mise-en-scène (e.g., stock characters), fair use (a huge can of worms), etc.
Re:Tencent Hunyuan-Large by ronsor on Tuesday November 05, 2024 @04:35PM (#42055381)
If the sole purpose of your model is to copy a work, then that's copyright infringement.
Re:Tencent Hunyuan-Large by roywiggins on Tuesday November 05, 2024 @04:38PM (#42055400)
Oh, in this case, the model can either reproduce the work exactly, or it can play tic-tac-toe depending on how you prompt it.
Re:Tencent Hunyuan-Large by ronsor on Tuesday November 05, 2024 @04:41PM (#42055422)
We can change "sole purpose" to "primary purpose", and I'd argue something that happens 50% of the time counts as a primary purpose.
Re:Tencent Hunyuan-Large by PeterStuer on Wednesday November 06, 2024 @03:39PM (#42068987)
If the sole purpose of your model is to copy a work, then there would be far easier, cheaper and more reliable techniques to achieve that.

Judge the output, not the system.