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

191–200 of 342 posts

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#191

Post by paxys »

Jimmc414 wrote: post #188
paxys wrote: post #184

Earlier quoted context omitted.

That's just the base concern with every single model regardless of where they sourced their data from. Garbage in, garbage out.

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)?

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#193

Post by commoner »

bloppe wrote: post #185
commoner wrote: post #175

Earlier quoted context omitted.

Also, section D3 of the GitHub Terms of Service says: > You retain ownership of and responsibility for Your Content. and section D4 says: > This license does not grant GitHub the right to sell Your Content. It also does not grant GitHub the right to otherwise distribute or use Your Content outside of our provision of the Service, except that as part of the right to archive Your Content, GitHub may permit our partners…

The clauses always have a trap door: "[outside of] our provision of the Service" means they can do anything as long as it's a service they provide. Under definitions: The “Service” refers to the applications, software, products, and services provided by GitHub, including any Beta Previews.

I think there's a misunderstanding over what the word "relinquish" means.

The terms make clear that uploading code to GitHub gives GitHub the right to "store, archive, parse, and display Your Content, and make incidental copies, as necessary to provide the Service, including improving the Service over time" while the code is hosted on GitHub.

However, that's not the same thing as relinquishing (giving up) licensing rights to GitHub. The uploader still retains those rights, and there is nothing in the terms that says otherwise.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#194

Post by int_19h »

blisterpeanuts wrote: post #87
century19 wrote: post #14

Google accused Microsoft Bing of using them for page rankings a few years ago. Setup a sting to show that when you searched for something unique on Google using MS Explorer, shortly afterwards the same search result would start showing up on Bing. This was seen as deeply embarrassing for Microsoft at the time.

Embarrassing, maybe, but imitation is the sincerest form of flattery.

Indeed, which is why the biggest impact this revelation is likely to have (if proven true) is on Google's stock.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#195

Post by abduhl »

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)?

Your argument would have a lot more force if we were past that point rather than fast approaching that point. Concerns about training data errors being compounded are much more important when you're talking about the bleeding edge.

And your question about how OpenAI prevents their training data from being corrupted is one we should be asking as well!

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#196

Post by Ifkaluva »

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.

Regarding point 2, I think there's nothing "wrong" with it, mainly it's funny that they don't know how to do it themselves. Provides additional evidence that Google is outgunned in this fight.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#197

Post by wendyshu »

mattbee wrote: post #6

Good luck to them. AI models are automated plagiarism, top to bottom. None of us gave OpenAI permission to derive their model from our writing, surely billions of dollars worth, but they took it anyway. Copyright hasn't caught up so all that stolen value rests securely with OpenAI. If we're not getting that back, I don't see why AI competitors should have any qualms about borrowing each others' work.

Is Stack Overflow plagiarism?

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#198

Post by shadowgovt »

Apart from the open questions of the quality of such once-removed-from-human-generated training data...

I can't speak to the legality of the situation, but the morality of using, without their consent, data generated by someone's AI engine...

... that was, itself, trained on other people's data without their consent...

... should be, at the very least, equivalently evil to the original AI's training.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#200

Post by sdenton4 »

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…

Things like ShareGPT or PromptHero give vast repositories of human-curated ML outputs, which make them fantastic for at least incremental improvement on the base model. In the grand scheme of things, these will be just another style, mixed in with all the other crap in the training set, so I don't imagine it's too harmful... eg, 'paint starry night in the style of midjourney 5'
Post reply on HN