Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google denies training Bard on ChatGPT chats from ShareGPT

twitter.com

Post reply on HN
Print view View original

291–300 of 342 posts

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#291

Post by staticman2 »

LegitShady wrote: post #224
ghostbrainalpha wrote: post #123

Earlier quoted context omitted.

How could that be ever really be enforceable? If I use an AI tool to design my Superhero, can't I just submit it without disclosing the help I received from an AI. I get that it would be very nice to prevent AI SPAM copyrighting of every possible superhero, but if I use the AI to come up with a concept, then quickly redraw it myself with pen and paper, I feel like it would never be provable that it came from an AI.

you would be committing fraud. what happens if a criminal commits fraud?

Redrawing something by hand creates a new copyrightable work, so it certainly isn't fraud to claim you own the copyright in a work of art you drew based on an AI output.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#292

Post by LegitShady »

staticman2 wrote: post #291
LegitShady wrote: post #224

Earlier quoted context omitted.

you would be committing fraud. what happens if a criminal commits fraud?

Redrawing something by hand creates a new copyrightable work, so it certainly isn't fraud to claim you own the copyright in a work of art you drew based on an AI output.

It depends if your redrawing is substantially different enough from the original image to earn copyright on its own. Your changes to an image from ChatGPT do not affect the copyrightability of the original content. If you've simply redrawn what the computer designed it may not be substantial enough to earn copyright. If you've made changes, it may only be copyrightable for those changes.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#293

Post by wdpk »

halfeatenscone wrote: post #95
wdpk wrote: post #56

even if true which it does not seem to be the case, the whole thing sounds pretty marginal, in order to train a model that is most likely significantly bigger than 100b parameters, one also needs orders of magnitude more training data than the small 120k chat that were shared on the ShareGPT website

Such logs would not be used for training the base model, but rather for fine-tuning the model for instruction following. Instruction tuning requires far less data than is needed for pre-training the foundation model. Stanford Alpaca showed surprisingly strong results from fine-tuning Meta's LLaMA model on just 52k ChatGPT-esque interactions ( https://crfm.stanford.edu/2023/03/13/alpaca.html ).

well, the initial twitter rant was pretty bombastic:

"The cat is finally out of the bag – Google relied heavily on @ShareGPT 's data when training Bard.

This was also why we took down ShareGPT's Explore page – which has over 112K shared conversations – last week.

Insanity."

Fine-tunning is not exactly the same as "relying heavily", I bet they got way more fine-tunning data from simply asking their 100k employees to pre-beta test for a couple of months

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#294

Post by shagie »

anotherman554 wrote: post #284
shagie wrote: post #279

Earlier quoted context omitted.

Your one sentence from one thousand works is likely seen as transformative. https://www.copyright.gov/fair-use/ > Additionally, “transformative” uses are more likely to be considered fair. Transformative uses are those that add something new, with a further purpose or different character, and do not substitute for the original use of the work. Creating a model from copyrighted works is likely sufficiently transformat…

"Creating a model from copyrighted works is likely sufficiently transformative to be non-infringing even if it is found to be a derivative work." Maybe, but one of the factors of fair use is whether it deprives the copyright owner of income or undermines a new or potential market for the copyrighted work. If ChatGPD gets so good at writing J.K. Rowling novels that it hurts the sales of the next J.K. Rowling book, tha…

But the model doesn't have any agency.

GPT isn't spitting out novels in the style of J.K. Rowling and sending them to publishers - a human is.

GPT being instructed to tell a Harry Potter story itself is no more infringing than a child asking a parent for a made up Harry Potter bed time story. They equally infringe and undermine new or potential markets for copyrighted work.

The question is "what do you do with the material?" If a human took the output of GPT writing as J.K. Rowling or a parent took their collected Harry Potter bedtime stories - those are equally problematic.

If I was to take a portrait of Marilyn Monroe and send it through a plugin called Warholize in Photoshop ( https://www.adobe.com/creativecloud/photography/hub/guides/c... ) , it's not the plugin or photoshop that is infringing - it would be me, the human who created an infringing work. If I print it out and hang it on my wall, that didn't particularly impact on the income for the Warhol estate nor deprive them of new markets. If I print out copies of it and sell them - then that is a different matter.

The question is what you - the human with agency - do with the infringing work after you create it. You can't blame photoshop for creating a Warhol infringing work nor can you blame GPT for writing in the style of J.K. Rowling if you instruct it to do so.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#295

Post by pessimizer »

Imnimo wrote: post #110

I don't care at all about this from a copyright or data ownership perspective, but I am a little skeptical that it's a good idea to be this incestuous with training data in the long run. It's one thing to do fine tuning or knowledge distillation for specialized domains or shrinking models. But if you're trying to train your own foundation model, is relying on output from other foundation models going to make them lea…

The internet is an easy, convenient way to train LLMs, but I'm pretty sure you could train them with microphones. One cloud surveillance company, like maybe for networked security monitoring, or maybe just Alexa/Siri etc. could dip into as many and as varied communications per hour than all the books ever written.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#296

Post by pessimizer »

CuriouslyC wrote: post #117
sebzim4500 wrote: post #94

Earlier quoted context omitted.

And how are you going to distinguish those interactions from chatbots trying to sell you something?

A network of trust, backed by a social graph, which can be used to filter untrusted content.

We've been pretending to be just about to do this for decades. The fact is that internet companies will not develop a network of trust, because they are primarily advertisers looking for better ways to abuse trust.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#297

Post by p1necone »

paxys wrote: post #191
Jimmc414 wrote: post #188

Earlier quoted context omitted.

Right, but training an LLM on the output of another LLM can certainly exacerbate these issues

Maybe, but we are fast approaching the point (or more likely have crossed it already) where distinguishing between human and AI generated data isn't really possible. If Google indexes a blog, how does it know whether it was written with AI assistance and therefore should not be used for training? Heck, how does OpenAI itself prevent such a feedback loop from its own output (or that of other LLMs)?

Pre-AI data is going to become like pre-nuclear steel.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#298

Post by staticman2 »

LegitShady wrote: post #292
staticman2 wrote: post #291

Earlier quoted context omitted.

Redrawing something by hand creates a new copyrightable work, so it certainly isn't fraud to claim you own the copyright in a work of art you drew based on an AI output.

It depends if your redrawing is substantially different enough from the original image to earn copyright on its own. Your changes to an image from ChatGPT do not affect the copyrightability of the original content. If you've simply redrawn what the computer designed it may not be substantial enough to earn copyright. If you've made changes, it may only be copyrightable for those changes.

The example was redrawing something by hand that was computer generated originally.

It would be pretty much impossible for a hand drawn work of art to not be sufficiently original. Hand drawn art doesn't look the same as what a computer produces. Originality has a very low threshold, simply pointing my camera at something and hitting click is almost always enough to show originality.

At any rate it isn't fraud to take the legal position that you are an original enough artist to have copyright in the work. If taking a legal position was "fraud" any attorney who lost a court motion would be whisked away to jail.

Edited to add: the copyright registration form asks if you are the "author" not if you are "original."

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#299

Post by b7r6 »

paxys wrote: post #88

1. Google denies doing it, so at the very least the title should have an "allegedly". 2. Even if they did – so what? The output from ChatGPT is not copyrightable by OpenAI. In fact it is OpenAI that is training its models on copyrighted data, pictures, code from all over the internet.

Whether or not training one giant LLM with socially enormous stakes on the output of another commercially-controlled LLM is an interesting question.

My stronger opinion is that the people who can do this stuff via having a crawled corpus of the Internet need to keep in mind that it's all our "user-generated content" that they've freely appropriated to build their models, and so whatever the technical copyright rules are (or become): you don't ethically own something that's closely imitating stuff we all wrote over the years.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#300

Post by layer8 »

asddubs wrote: post #211
layer8 wrote: post #186

Earlier quoted context omitted.

Two wrongs don’t make a right.

forgive me if i have limited sympathy when a burglars house gets robbed

That's not how rights work, though. I'm sure you don't want your rights to be conditional on the sympathy of others.
Post reply on HN