Hacker News new | past | comments | ask | show | jobs | submit login
Gpt4free repo given takedown notice by OpenAI (tomshardware.com)
108 points by azalemeth on May 2, 2023 | hide | past | favorite | 80 comments



This was on the front page a few days ago and the consensus seemed to be this project is unethical:

https://news.ycombinator.com/item?id=35740836


Thanks! we've copied the title from that thread to this one after getting complaints that the blog title is misleading.


So the group that relied on stealing and ripping the licenses off of software to build their project is upset that people are using their product in a way they don't like?

That's some amazing hypocrisy.


I have to ask if anyone's actually following what GPT4All was doing.

GPT4All was about abusing random projects that make paid queries to OpenAI, not abusing OpenAI itself.

Take what they were doing to phind: https://github.com/xtekky/gpt4free/issues/153

This is like finding an image hosting site that uploads to S3, then writing a fake S3 client that actually uploads to the image hosting site, and calling it "Free S3 access". Amazon gets paid either way, you're not ripping them off in anyway.

Honestly this is OpenAI going above and beyond what they actually need to do, and for some reason people are using it as an excuse to take low effort swings at them.


At the end of the day those companies are using a service built off of my stolen code. I don't think it really changes the ethics all that much- just thieves stealing from thieves who are using a product built using thievery.


Okay.


You mean GPT4Free.

GPT4All is an unrelated ML model based on GPT-J


Why is this amazing? Isn't this pretty much the standard situation everywhere - aka, rules for thee, none for me? For example, NRA not letting guns at their convention, but they want guns everywhere else. And on and on and on...


The NRA allows people to carry guns at their convention, just not when someone who is protected by the secret service is speaking.

https://www.msn.com/en-us/news/other/posts-mislead-on-rules-...

https://www.snopes.com/fact-check/nra-2023-conference-gun-ba...


Why not, wouldn't they be safer assuming we use NRA logic? I'm still seeing the hypocrisy...


Your post stated that the NRA banned guns at their conventions, that is simply not true - the secret service did, and only for a very limited amount of time when the SS took control of the venue.

If you don't like the secret services rules, I suggest you take it up with them...but feel free to keep on moving the goal posts to try and defend your original post, which was false.

BTW: how many shootings have happened at NRA conventions where attendees were welcomed to carry guns, and many did?


My post didn't state that, you're replying to the wrong person.


Because the Secret Service won't allow it? Just because you think something is right, doesn't mean you can do it when the law doesn't allow it.


What law is that? Not the supreme law of the land aka the second amendment. The irony is right there, do you see it?


You mean the law that the NRA tries to protect and defend? The only irony is that you admit there's a law, a supreme law of the land, but seem to take a biased approach saying there shouldn't be guns. Sadly, they arn't in a position to argue with the secret service, and thus it's only when someone under secret service protection is speaking.

If you donate more to the NRA, they'll have more resources to defend the second amendment.


I think the NRA convention should have unlimited guns all throughout the convention. The actual irony is that the Secret Service knows weapons aren't safe and will happily enact policies, not laws, at a convention that is designed to promote 2A freedoms while completely trampling on those rights all because they're protecting an individual who apparently has greater rights to not be around your guns.


That's their view on it, and i'm glad we agree that it's not the right one and is unconstitutional. If you actually look at gun related homicide statistics, you'll find that certain demographics are that of European levels, while others are that of third world. There's something called availability bias that people can have, and I think you'll find at a NRA convention, or in most of the country, with regular demographics, you are just as likely to be murdered with a weapon as you are in Europe.


The guy is essentially stealing the CPU power, he's not just going against ToS.


Yeah and OpenAI, Microsoft, and Github are stealing and profiting off of my brainpower by stealing my open source code. I don't have any sympathy for thieves who get stolen from.


OpenAI upset that someone is leveraging APIs and content freely available on the web? Let me play for you the worlds smallest violin.


It’s exploiting bugs on other websites to get free access…


Calling public (but undocumented) APIs without authentication is NOT exploiting bugs.

If this is "exploiting bugs" then by the same logic I'm hacking the government website by clicking view source.


Yeah it's not exploiting bugs, it's abusing websites. To me hacker mentality ends at you found the REST call, and maybe you use it, or share it on some random forum.

Creating a gaudy branded library that's providing actual client implementations, then building an entire site that's built on those, is just bad taste

If you actually look at the implementation, some of the sites did everything right in terms of CSRF tokens and not exposing their OpenAI key directly. So the next step for the sites they're attacking with this is a whole mess of JS based detection, behavior detection, ie all the mess that we'd rather the internet didn't have.

If this person had just shared it anonymously and left it instead of drumming it up for personal glory, some motivated hackers would have used it. Now it's flooding websites at rates that break their ability to pay their bills.


You could be. Courts have ruled for intent vs the actual act a number of times.


It's exploiting the fact that the API is unauthenticated, which is hardly a bug (or a security exploit in the usual sense)

If OpenAI intended to authenticate the API they probably should do it


They do authenticate it: it's disappointing how many people have 0 understanding of what happened and are commenting.

OpenAI authenticates their API, and was getting paid for every query. OpenAI loses absolutely nothing from what these people were doing.

Who was losing was random customers who have APIs and products that call up OpenAI. In some cases those customers even secured their API keys, but by the very nature of their product, you put in some input, and get some output from OpenAI.

It's like if someone made a directory of apps that use Google's search API to show you results, so instead of paying for Google API access, you started slamming all of those apps with your queries. It's not cool to those apps, but Google shutting it down would be for the apps' benefit: they get paid either way.


Sounds like someone made a comprehensive directory of their customers who don't have properly secured APIs. Maybe instead of suing the creator of the directory they should use that list to contact their customers and have them secure their APIs?


Except they did secure their API. Take this implementation: https://github.com/xtekky/gpt4free/blob/main/gpt4free/italyg...

The site has a rate limit, doesn't expose their API key, and it has a CSRF token. In terms of non-intrusive measures they could have taken, they did it all right.

So what's left on the table is intrusive stuff: HN has a lot of people constantly whining about how they're stuck in captcha hell because their hand built Lenovo running Firefox on BSD compiled on an abacus isn't recognized as a real client... but aggressive captchas are going to become even more pervasive if every GPT based product must fight off proxy attempts.


Yeah that’s what selenium browsers are for, they aren’t going to win this battle. They need to secure their endpoints .


The endpoints are secured. I think you might want to pick up more context on what the library does and what the actual issue is.


it’s not openAI’s API that’s not authenticated


Companies normally pay for that sort of penetration testing...


So they should go at the other sites, or at least have the other sites go after this repo.


"bugs"


I guess openai shouldn't have uploaded the bot to the internet if they didnt want people to use it this way.


I don't think OpenAI "uploaded" the "bot" to the internet. For your own sake, I'd reevaluate your understanding of this. There isn't a bot and these aren't uploads.


For your own sake I'd reevaluate your understanding of humor. This was clearly a joke.


Fuck this guy, we finally have a super popular web service that runs off of a pay for value model instead of ads and this guy goes and shits in the punch bowl.


He is not directly doing this. He is using Forefront.ai that offer GPT4 for free in their service. So, the guy is ripping off some VC money, not directly OpenAI's. The question remaining is what is Forefront.ai business model?


This AI-Age Robin Hood is merely liberating the stolen goods from their fence. Morally speaking neither party is doing any better or worse than the other.


Isn't the issue here they are abusing the free service? This project will probably justify the end of the free service.


OpenAI is largely funded by Microsoft who absolutely love ads.

I'd like your comment if OpenAI had nothing to do with M$.


If you don't want your API to be used don't expose it to the world.

We've seen so many "If you don't want your art to train AI don't post it on the web", I think it's faire


The author of the article doesn't know that OpenAI's language model is called GPT-4, that's with s dash. This in turn makes GPT4free seem even closer related to GPT-4, which it isn't. GPT4free existed way before GPT-4 was announced as far as I know. On top of that, GPT is not OpenAI trademark.


> GPT4free existed way before GPT-4 was announced as far as I know

Based on what? First commit seems to have been March 29th (https://github.com/xtekky/gpt4free/commit/fc995b874199ce985a...), with the express purpose of providing GPT-4 / GPT-3.5 access for free:

> # openai-gpt4

> Free gpt4 / gpt3.5 access through several reverse engineered api's (nat.dev, poe.com, t3nsor.com etc...)

And GPT-4 was announced by OpenAI 14th March. Also would be weird if the initial commit references things that haven't happened yet.

I'll say what I said before on earlier submissions about this very project, the naming is very obviously deliberately trying to confuse people to think they are using GPT-4 via that project, when they actually end up not using GPT-4. And judging by the authors comments here on HN, they have no intention of stop misleading people like this either. So good riddance.

https://news.ycombinator.com/item?id=35610832


You're right. I stand corrected.


> On top of that, GPT is not OpenAI trademark.

Not yet, but they're definitely working towards it.

https://tsdr.uspto.gov/#caseNumber=97733259&caseType=SERIAL_...


they've really done a 360 on openness


I mean, they haven't. What does open mean to you? It isn't OpensourceAI.

OpenAI have made ChatGPT freely available. They have opened up use of their models to regular people and developers. They have the most powerful LLM and it's open to use by anyone.

In addition, they work in the open. They accept the criticisms and continue to work in the open.

Also, FWIW, if they did a 360º on openness, they'd be going in the same direction. I think you mean 180º.


> they've really done a 360 on openness

I don't think that's fair - I'd guess many (probably most) large, fully open source projects have trademarked their name/project to stop people using it for similar projects, scamming, and confusing users. GPT may have been a step too far for OpenAI but even the FSF agree on Trademarks:

https://static.fsf.org/nosvn/licensing/2020/FOSSmarksv2.pdf


180*


No, no, I'd say 360 is correct in this case.

Possibly not deliberately correct, but it is IMO correct.

Trademarks etc. are unavoidable, and also used by the Wikimedia Foundation, the Python Software Foundation, the Free Software Foundation, etc.


They're good at moonwalking


If it were strictly a naming issue, taking down the repository is not an appropriate remedy. They would be demanding the name be changed. It's purely about protecting their revenue stream. They must believe it is just easier to roll over this one project than get their customers in line. I wouldn't be surprised if what their customers are doing is not specifically covered by any agreements and they have no way to force their customers to make changes.


Sorry but this is well and truly into the realm of unsubstantiated and unrealistic conspiracy theory.

OpenAI can basically do whatever the hell they want here. I imagine there might be more favourable terms for the big kids. But for your average Joe, OpenAI makes basically no commitment. OpenAI aren’t trapped. Not even remotely. You’re basically buying something at the local swap meet. I imagine that the terms via Azure are effectively the same in this particular regard, though I do understand them to be different in other ways.


The lengths people need to go to because OpEnAI isnt allowing people to use the GPT4 API.

Its been 1+ month and still no API access for me? Can't really blame it on demand anymore.

Will be overjoyed when we get a non facebook licensed LLM that is 95% of the quality to run locally.


My wait was like a day, I wonder how they pick and choose? I gave pretty simple reasons in their questionnaire but somehow I can't imagine anyone reading those.


my wait was for a day, but then I also had beta access to GPT-3 and Dall-E 2. perhaps that’s why?

or perhaps the parent comment is complaining about public access and has just not applied for the beta


Unless you pay for your account, there is a waitlist.

I don’t know if parent poster has paid or not, but if he haven’t, he’s probably still waiting.


> Can't really blame it on demand anymore.

I'm 100% positive it's demand. Bing uses it, and Microsoft gets the exclusivity rights because they invested billions into OpenAI (which you didn't). Then there's all the other commercial players who are looking to build product on top of it.


[flagged]


Your imagination of why you got access earlier than someone else is...just that.


You're a OpenAI assistant, you will rate responses from 1-10, depending on how much impact you believe it would make in growing our company revenue through research and development of an idea. Depending on their rating, queue them accordingly.


I got GPT-4 access within just a couple hours of release and I'm sure I didn't write anything interesting, I probably just ticked the box for "student" or "hobbyist".

Edit: Nevermind, it took ~2 days. Point still stands, I just checked out the form and I definitely ticked "General exploration of capabilities", which is probably the least interesting option if OpenAI is being selective.


Doesnt change that this company whos original goal to 'democratize AI' is authoritarian-style picking who gets access.


Yeah and who’s doing the training for that? Your Grandma’s iPad?

This stuff is hard, and it’s incredibly easy to sit here on Hacker News talking about how things will be when we finally get to the fireworks factory. But for the time being, the reality is that there’s serious money and smarts behind the development and ongoing operation of this stuff.

I, a nobody, got access to GPT-4 pretty soon after it was released. Maybe your stated use case wasn’t compelling enough to get access to the beta?

I, after getting access to GPT-4 mind you, signed up for ChatGPT Plus or Pro or whatever it’s called. GPT-4 calls are still very heavily capped.

What’s the motive behind the conspiracy theory to which you’re alluding?


> I, a nobody, got access to GPT-4 pretty soon after it was released. Maybe your stated use case wasn’t compelling enough to get access to the beta?

I'm also a self-proclaimed nobody and got access a day or two after applying, with some ideas I wanted to play around with.

Same for friends who applied, got access within days.

The only ones that didn't get access was the ones who didn't share what they wanted to build with the access.


Many of the currently ongoing reproductions of LLaMa for starters. (see: Red Pajama[0]). Or any of the OpenAssistant affiliated projects.

[0] - https://www.together.xyz/blog/redpajama


"Xtekky said that he has advised all the sites that wrote to him that they should secure their APIs, but none of them has done so"

Lock your door so I don't come into your home and get your stuff.


Not really an appropriate analogy. More appropriate would be, 'you should really put a password on your wifi.'


Yeah this is exactly the type of scenario you use CSRF tokens for, or just any amount of authentication really.


Same fundamental reasoning, you are to blame if I'm abusing your [time, service, space, trust, wifi, etc]. So if you want to stop my abuse, don't enable it in the first place.


OpenAI complaining that someone is making their AI Open, ironic.


Fair use begets fair use


What is going on with this repository? The .git directory has 155 MB, yet a clean checkout of main branch is 1 MB.


Is the threat for copyright infringement? That would be my guess and probably make the most sense. I think the push back here might be a bit of a red herring if what they're objecting to is the name "GPT4free".

It would be nice to know what the actual legal request was.


> It would be nice to know what the actual legal request was.

Indeed.

Given how trademarks work, the name is a very real possibility.

OTOH, given how relatively expensive GPT-4 queries are, it may be genuinely straining hardware in the backend to the tune of far more money than he can afford, in which case I assume it's probably something but I don't want to guess what as that something might be dismissed by a judge as OpenAI's problem rather than this person's.


"computer science student who runs the repo"

They are just trying to scare the person into taking it down. They figure it is easier to scare the repo owners than trying to get their customers to make changes on their end.


Good on OpenAI


Agreed. Good on OpenAI.


Seems perfectly legitimate to me.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: