204
this post was submitted on 05 Nov 2023
204 points (93.2% liked)
Technology
59308 readers
4864 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
This might upset you but for some uncensored model that have alignment removed will output such content. Is the content true? Don't know cuz I haven't read Harry Potter.
Sure, then whoever uses it to extract that text is infringing. If I memorize a copyrighted text, my brain is not an infringement; but if I publicly perform a recitation of that text, that act is infringing.
Really the precedent of search engines (and card catalogs and concordances before them) should yield the same result. Building an index of information about copyrighted works is librarianship; running off new copies of those works is infringement.
On the other hand, AI transparency is also an interesting problem. It may be that one day we can look at a set of neural network weights -- or a human brain! -- and say "these patterns here are where this system memorized Ginsberg's 'Kaddish'." I hope we will not conclude that brains must be lobotomized to remove copyrighted memorized texts.
If we treat model like a brain that "memorize" copyrighted text and generate new text based on that, your statement is valid. However, this will also prohibit any copyright claims on the model's output, as the act of memorization isn't a work. Only work can infringe on other works, which should the output of models defined as "work" is still under heavy debate. Even if it is defined as a work, can a model gain copyright while not being a legal person? Who should bear the liability then? What if the output is modify by an editor? This rabbit hole digs deep.
I think that actually was ruled on a few months ago. No the model cannot hold copyright. Nor can the person that commissioned the model to create the work. I think where things are still a bit grey (someone correct me if I'm wrong), is when a person creates a work with the assistance of AI whereas it's a mix of human and AI generated content.
The model doesn’t contain the training data—it can’t reproduce the original work even if it were instructed to, except by accident. And it wouldn’t know it had done so unless it were checked by some external process that had access to the original.
In case anyone wants to try this out: Get ComfyUI and this plugin to get access to unsampling. Unsample to the full number of steps you're using, and use a cfg=1 for both sampler and unsampler. Use the same positive and negative prompt for both sampler and unsampler (empty works fine, or maybe throw BLIP at it). For A1111: alternative img2img, only heard of it never used it.
What unsampling is doing is finding the noise that will generate a specific image, and it will find noises that you can't even get through the usual interface (because there's more possible latent images than noise seeds). Cfg=1 given the best reproduction possible. In short: The whole thing shows how well a model can generate a replica of something by making sure it gets maximally lucky.
This will work very well if the image you're unsampling was generated by the model you're using to unsample and regenerate it, it will work quite well with related models, imparting its own biases on it, and it's way worse for anything else. If you ask it to re-create some random photograph it's going to have its own spin on it changing up pretty much all of the details, if you try to do something like re-creating a page of text it's going to fail miserably as stable diffusion just can't hack glyphs.