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

161–170 of 342 posts

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#161

Post by dragonwriter »

seanhunter wrote: post #156
sebzim4500 wrote: post #149

Earlier quoted context omitted.

No one is alleging copyright violations. The claim is that they violated OpenAI's terms of service. We don't know whether Google ever even agreed to those terms of service in the first place.

Are OpenAI saying they have adhered to the terms of service of all the content they have used?

Content is not subject to terms of service.

Services are subject to terms of service. (If content is received through a service, the terms of service may govern use of it, but that’s not a feature of the content, but the acquisition route.)

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#162

Post by bloppe »

dragonwriter wrote: post #143
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.

> The output from ChatGPT is not copyrightable by OpenAI. I think the argument here is over the OpenAI Terms of Service, not copyright.

See HiQ vs LinkedIn. ToS has to be affirmatively accepted. I doubt that happened in this case.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#163

Post by flir »

bloppe wrote: post #158
saurik wrote: post #140

Earlier quoted context omitted.

So, to verify, are you claiming that if someone added a similar clause to their source code and then GitHub went ahead and trained Copilot against it, that would be an issue?

You relinquish all licensing rights when you upload your code to GitHub. Microsoft can do whatever they want with it. That's in their ToS, which you have to agree to when you make an account. Normally, only affirmatively accepted ToS are enforceable, so just putting a clause into your license doesn't work (unless it's a copyright, which doesn't require consent).

> You relinquish all licensing rights when you upload your code to GitHub

What now? Seriously?

I found this. Section D4.

"We need the legal right to do things like host Your Content, publish it, and share it. You grant us and our legal successors 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. This license includes the right to do things like copy it to our database and make backups; show it to you and other users; parse it into a search index or otherwise analyze it on our servers; share it with other users; and perform it, in case Your Content is something like music or video."

"as necessary to provide the Service" seems critical.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#164

Post by dragonwriter »

paxys wrote: post #160
dragonwriter wrote: post #143

Earlier quoted context omitted.

> The output from ChatGPT is not copyrightable by OpenAI. I think the argument here is over the OpenAI Terms of Service, not copyright.

And what about the terms of service of my blog or code repository? Does OpenAI respect that?

> And what about the terms of service of my blog or code repository? Does OpenAI respect that?

Seems to me that’s an issue between you and OpenAI. (Does your blog or code repository actually have published restrictive terms of service? Did it when OpenAI accessed it? Did OpenAI even access it?)

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#165

Post by paxys »

vagabund wrote: post #148
KRAKRISMOTT wrote: post #134

Earlier quoted context omitted.

OpenAI Terms of service forbid training competitor models via their ML outputs (LoRa alpaca laundering is probably not allowed for commercial use).

Google has no contract with OpenAI though. They used a third party site to scrape conversations. If the outputs themselves are not copyrighted, and they never agreed to the terms of service, it should be fine, right? Albeit unethical and embarrassing.

Hardly unethical, considering OpenAI is doing exactly this.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#166

Post by Jimmc414 »

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.

>Even if they did – so what?

Amplification of biases, propagation of errors, echolalia and over-optimization, lack of diverse data, overfitting

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#167

Post by danShumway »

magicalist wrote: post #145
danShumway wrote: post #55

So? First off, the whole argument behind these models has been from day one that training on copyrighted material is fair use. At most this would be a TOS violation. Second off, AI output is not subject to copyright, so it has even less protection than the original works it was trained on. Copyright maximalism for me, but not for thee. It's just so silly for someone working at OpenAI to complain about this.

> At most this would be a TOS violation And would it be a ShareGPT TOS violation (assuming it had any)? If OpenAI says "you can share these online but don't use them for AI training", people share them on another site, and then someone else comes along to scrape that site for AI training data, there's no relationship between OpenAI and the scraper for the TOS to apply to. Normally I think you'd rely on copyright in t…

Right. And what even is the penalty of that TOS violation and how enforceable is it?

I don't have an OpenAI account. I have never agreed to any TOS. I don't see what legal claim they would have to stop me from training an LLM on ShareGPT.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#168

Post by brucethemoose2 »

sebzim4500 wrote: post #154
brucethemoose2 wrote: post #137

This is also bad because the risk of AI "inbreeding" is real. I have seen invisible artifact amplification happen in a single generation training ESRGAN on itself. Maybe it wont happen in a single LLM generation, but perhaps gen 3 or 5 will start having really weird speech patterns or hallucinations because of this.

Worst case scenario they just start only training on pre-2020 data and then finetuning on a dataset which they somehow know to be 'clean'. In practice though I doubt that AI contamination is actually a problem. Otherwise how would e.g. AlphaZero work so well (which is effectively only trained on its own data).

It could degrade the model in a way that avoids the metrics they use for gauging quality.

The distortions that showed up in ESRGAN (for instance) didnt seem to effect the SSIM or anything (and in fact it was training with MS SSIM loss), but the "noise splotches" and "swirlies" as I call them were noticable in some of the output, but you have to go back and look really hard at the initial dataset to spot what it was picking up. Sometimes, even after cleaning, it felt like what it was picking up on was completely invisible.

TLDR Google may not even notice the inbreeding until its already a large issue, and they may be reluctant to scrap so much work on the model.

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#169

Post by saurik »

bloppe wrote: post #158
saurik wrote: post #140

Earlier quoted context omitted.

So, to verify, are you claiming that if someone added a similar clause to their source code and then GitHub went ahead and trained Copilot against it, that would be an issue?

You relinquish all licensing rights when you upload your code to GitHub. Microsoft can do whatever they want with it. That's in their ToS, which you have to agree to when you make an account. Normally, only affirmatively accepted ToS are enforceable, so just putting a clause into your license doesn't work (unless it's a copyright, which doesn't require consent).

So, to verify, are you claiming it would not be allowed for you to upload my otherwise-open-source code (code I do not myself host at GitHub, but which was reasonably popular / important code) to GitHub?

Re: Google denies training Bard on ChatGPT chats from ShareGPT

#170

Post by kweingar »

vagabund wrote: post #148
KRAKRISMOTT wrote: post #134

Earlier quoted context omitted.

OpenAI Terms of service forbid training competitor models via their ML outputs (LoRa alpaca laundering is probably not allowed for commercial use).

Google has no contract with OpenAI though. They used a third party site to scrape conversations. If the outputs themselves are not copyrighted, and they never agreed to the terms of service, it should be fine, right? Albeit unethical and embarrassing.

> Albeit unethical and embarrassing.

I really don’t understand this angle. In fact, I am fairly positive that the training set for GPT-4 contains many thousands of conversations with AI agents not developed by OpenAI.

Do AI companies need to manually sift through the corpus and scrub webpages that contain competitor LLM output?

(“Yes” is an acceptable answer to this, but then it applies to OpenAI’s currently existing models just as much as to Bard)

Post reply on HN