chats disappeared #3695
Comments
|
I observed the same. |
|
im observing the same problem. i wonder what is the reason behind |
|
I can still load https://open-assistant.io/chat and get the static content, but it seem not to be able to load data. Request to URL https://open-assistant.io/api/chat?include_hidden=false fails (404). Maybe the backend is offline?
|
|
3 days without any answer from admins. We know that you don't owe us anything, that the project is being offered free of charge, and so on. But we also spend time labelling content, correcting answers, and we'd like to continue using the chat while offering you our collaboration. We're clearly hoping that this isn't the case, that we're just kidding ourselves and that there's just a communication problem. And that the interface bug has just not been brought to your attention in time. However, on Discord and here, many are concerned... |
|
You don't suppose they let us use open assistant for free intially and then disappearing with community work? |
I don't think we should jump to conclusions, maybe it just needs a fix. Have they said anything about this? |
The fact that no one is responding after several days, either on Discord or in the GitHub issues, may lead many of us to think this kind of thing. It wouldn't be the first time a project decided to do something stupid like this. My guess is just an expression of a fear that was shared on Discord (not by me), and I think it highlights that lack of communication during a service outage can lead to this kind of talk. So please don't shoot the messenger, understand him, and respond diplomatically. |
Andreas talked in several podcasts about the fact that the heavy unpaid workload isn't sustainable and can't go on for ever. Maybe the team just has different things to do right now in their private life. Give them at least a few weeks to react, this is still an open source project and everyone is doing this in their free time. |
A simple message on the web page "we have some points to fix" takes a few minutes. Answer here, labeling "in progress" is a simple click. There is a team, there are several managers. The website is used by thousands of users / contributors. Of course, it's free. But the communication is important in this case. |
|
I just want to give a precision. I just explained on Mastodon that there are many open source projects to use AI with respect of privacy. Then, I gave open assistant as a "good example". The answers were "LOL, great, there is no chat"... I cannot defend the project now... Because I don't have explanation to give to the bad comments, and I cannot defend that it works... The problem is that the "open source" image is impacted when no one gives a status on outage. |
|
I understand your frustration and this actually happens very often with open source projects. I think they should be clearer about the fact that the website is more a proof of concept and a tool to collect the dataset. But it is not meant to be a website that is used by millions of people. I actually would prefer a paid model by a third company who hosts the OA model. |
|
I can't imagine how there could be any good reason for no word at all in four days, but maybe there is. Doesn't anyone have any way to contact any of the developers or admins? |
|
Guess political agendas did their thing and now Open Assistant is forbidden/banned. When will they stop this nonsense censorship? They want to spread the message that AI is dangerous and "stop" it because they want it only for themselves to control - to use it to their advantage for their political purposes, including censorship, disinformation and brainwashing. That explains the silence as well. Who knows what have they been threatened with, should they keep it open for everyone to use, or even say a word on why they closed access. |
So that they can make it "Closed Assistant" and do the same kind of thing OpenAI has done with Dall-e 2, forbidding certain subjects that they deem censorable (like Ukraine) or silently modifying prompts to forcefully add "diversity" to the output images you're paying for (such as gender, nationality, skin tone, sexual orientation, etc) ? To be honest, I would much rather not let Open Assistant become a tool to enforce political agenda and give this agenda (in incognito/undercover ways) a false sense of "legitimacy" that it doesn't really have in our society. The moment you take an Open AI (a truly open one), give it to a company and make it commercial, that moment the AI is going to be used by a few powerful people against the rest of society, for their own benefit and increasing their power even more. The whole point of Open AI software is opposite to taking control away from everyone, forbidding, censoring, purposely biasing the AI to purposely biase society through psychological manipulation, and in the end taking away everyone's freedom because it's up to the ones with money and power what others can or can't do, say or even think. The whole point of Open AI is that AI shouldn't be restricted to a few people to rule over the rest using this technology, money and several other means, which combined enslave everyone else without possible escape. Is that the world you want? |
|
Calm down, the models are freely downloadable. Everyone can run them or even change them. Right now Open Assistant is completely depended on donated GPU time from big companies. A payed service (or more than one from different countries) would grantee independence. The free website should stay available, but we can never expect it to be a reliable service without proper funding.
Yeah just ask them on Discord |
|
They aren't responding on Discord either, but maybe we could find some contact information on the laion.ai website, for people who may or may not answer questions about what's happening. |
That's right, because everyone has a computer with 24 GB+ of VRAM in their home. And no, splitting it between 2 or 3 companies would not guarantee independence. Censorship has no mercy and enough money (or other means such as violence, be it legal or illegal) will ALWAYS silence any company or person's voice. Aren't you familiar with the new legislations that will, in the future, determine how to forbid our access to AI? Maybe you should read some news. The fact that they haven't said a word speaks for itself, I hope to be wrong but they probably had the safety of close relatives threatened, hence the permanent silence and "no signs of life". |
|
They literately released a new German model today: https://laion.ai/blog/leo-lm/ |
And who is running it? Where's it available to use? If no one can run it, it's just a file with a lot of unusable data. |
|
It is in the first paragraph of the article I am out of this discussion here. |
Come back whenever Open Assistant is available... Everyone's waiting...... And then, you tell me how wrong I was ;) |
|
There was an announcement about it on Discord, saying that the chat has been disabled for now, and there will be more updates on it soon. Maybe I should close this issue, because it isn't a bug. |
Thanks for the greatly detailed "explanation" as to what has happened. It really cleared things up. Let's do as he suggested and let's all self-host the models on our 24+ GB GPUs computers that we all have at home. Also nice to see how only the chat is disabled, because it's okay if the chat's not available, but it's not acceptable if people stop training it while it's disabled (and with no word on it ever coming back)
I second this. In the end, ironically, community work will be bought and used against that very community that believed in open availability and usage of AI, because censorship and agendas don't care about the community, and with enough money and power no one is willing to keep everything open, and would even sell their own children if the offer (or the risk after refusing) is high enough. The idea behind open source AI, working together, etc, is all very romantic, but no one is willing to defend that idea as soon as their personal gains or loses have more priority for them than these romantic, utopian ideals. And defending censorship instead of fighting it, yes it may save your personal arse in the short term, but in the long term it will not only fail to save your personal arse, but also become a problem for the whole of society, just because someone made the wrong selfish/coward decision. It has been normalized in our society that social networks will collect and literally sell our data, people are aware of it, think that this is normal, and no one is doing anything about it. We have seen how this data can be used to manipulate opinions or votes. I suggest that you begin considering the implications that AI will be misused in ways like this, and that people will also eventually normalize that "AI belongs to big corporations and governments, not to everyone". In this scenario, abusive targeted advertising after a private phone call or having ideologies imposed and accepted as facts will be the least of our problems. |
|
Hey, let's think realisitically. We are all open source developers, no one affiliated get's paid any money to do anything. We all currently clearly depend on sponsorships of third party companies. You may say let's host the models in our 24+ GB GPUs, but we don't have them because they're expensive, right? Ever thought about how expensive it may be for those companies to host our Open Assistant models with a heap of people using it simultaneously? Maybe they got sick of investing their compute on us... Now I'm just saying my opinion and thoughts on this, and I'm sure that the leadership team for OA is working hard in the background to resolve the issue. We'll just have to wait and see what their updates will be about. 😉 |
This is exactly what I'm talking about. No one is doing this for charity or because they want everyone to have access to the AI. Instead, some big third party companies are taking control of everything by sponsoring the project, and the continuity of that sponsoring (and thus the project) depends on whether those third party companies' interests are met or not. Which means you are all subject to what those third party companies want, or you all have no money or computing resources required and cannot continue. In other words, you have to obey to your masters as if you were dogs, and if censorship or any other misuse is the way to go, so be it, because you depend on keeping them happy to have funding. And the day they decide they used "their dogs" all they needed, that day they will take all the technology, the data, etc, all for themselves and their own interests, and all of you will only be able to keep saying "but you can still self-host the models!", and repeat it as many times as you want, you don't have the money or resources anymore and it's up to the ones that have it to decide what to use that technology and power for. You're free to live in your bubble that you're being sponsored for charity, but that's a child's way of (not) understanding how this world and the people on it work, or why they do the things they do, such as sponsoring "evil AI technologies" (that they condemn) while meaning no harm for anyone at the same time... Or so you want to believe. |
|
I work for this project because I want to create an open dataset to train language models and because I want to have models that I can use on my own GPU cloud provider and hopefully in the future (when technology progresses) on a local machine. And that is exactly what I am getting. As long as they publish their models and their datasets under an open licence I am happy. The Chat on the website is a nice extra, but it was always clear that it won't stay for ever. What exactly do you expect from this project? I have the feeling that you expect them to start some kind of revolution or something. |
I see...
Will you also be happy when legislation is made so you your GPU cloud provider has to prevent you from using it to run AI?
Not at all. I wish for AI to be free for everyone to use and run, not for big companies and governments to take it for themselves and use it on their own benefit and against the rest of society. Making it illegal for others to have access or run it since it would take away this unfair advantage from them for political and commercial purposes. But you live in the fantasy that they will let everyone with a computer use AI, and miss the chance to get this unfair advantage against us. I don't expect your fantasy to be our real future, given the obvious nonstop effort to force the "AI is intrinsically evil" message into society (which is silently followed by an hypocritical "but it's not dangerous if we are the ones who use it") . Are you going to deny it now? And maybe Open Assistant being gone "for no reason" isn't some mystic coincidence, but one of many steps to take AI away from the ones they want to censor, oppress and chain with that very same AI software. Because being spied from your phone's microphone, social networks collecting and selling people's data etc is only the tip of the iceberg when you add the AI into it. I don't expect any "revolution". But I don't want a world where the likes of Biden, Zuckerberg, Bezos, take the AI for themselves and with their computational power and massive amounts of data decide what people can or can't say, do or even think, starting by illegalizing AI much like Italy, but a general ban this time. It is common knowledge that privacy on the internet is but an illusion. Now add it all together and see what follows. Not that I have "the magical solution", but the more AI we can run locally, the more we're restricting their chances of controlling us, since they have to basically take away computers from everyone or imprison everyone to prevent AI usage, which fortunately to us are not realistic options for them. As soon as you have to rely on a "GPU cloud provider" though, you're screwed. |
|
Fwiw anyone can now deploy their own LLM for free with Cloudflare's WebGPUs - https://blog.cloudflare.com/webgpu-in-workers/ |
|
This issue is resolved for me, but I'm not sure if I should be the one to close it, or not. |
The fact you can do it now doesn't mean you'll be able to do it in the future. Cloudflare isn't exempt from having to comply with legislation and regulations. Yes, those don't forbid AI now, but just give them time.
I wouldn't say that it's resolved when:
It looks like they aren't going to say anything or resume the project (other than keeping collecting data while they still can). So I guess if we're waiting for eternity here, we might as well accept what has most probably happened and move on to somewhere else. RIP. Guess it couldn't have been any other way. It's just sad to see how everyone always surrenders to money and power in one way or another. How we have found cures to otherwise fatal health problems, or created the greatest technologies, and at the same time our freedom is not in our own hands but in the hands of a few. Freedom of expression, freedom of information, freedom of thought, freedom of choosing how to live. Everything is taken away from us, and the few "shields" that we have to defend ourselves are less and less with every day that passes. Ironic how humans use their intelligence in such ultimately unintelligent ways, because there are no true winners when the path is censorship and supremacy. Even surreal how there are people listening to what they say on TV or the media, that "this is the end of an economical crisis", when that's just a fairy tale for elementary school children, since anyone that has made a purchase in the last months knows how idiotic this storytelling is, and that our problems are not ending but have just begun. |
|
I still don't know if I should be the one to close this, or not. It looks like a dead issue to me. If we haven't got an answer by now, we never will. |
Censorship won once again. RIP Open Assistant. |
|
I'm wondering what "open" means, if it doesn't include transparency and open, honest communication. |
The "open" in open assistant is the same as the "open" in OpenAI. The very exact same. |
The thing is they have told us the problem in the discord, granted it took them a while but in context it makes sense. |
|
Someone paid or coerce the previous investor to pull out funding the project, so that some other investor with their own woke or controlling agenda and censorship will come and take over. open assistant RIP. |
I've already explained this, if it was as easy as a funding problem, the training from community work would also have become unavailable, which wasn't the case. It's hypocritical from you to try to mask/disguise the obvious while attacking OpenAI. Also, the only Discord message didn't explain anything: Hi everyone, we've disabled chat on the website for now. Soon, when? 7 days soon?
Thanks, you were able to sum it up much better than me, indeed I'm bad at summarizing so my explanations were much longer, but saying the exact same because that's how it is and it's idiotic to deny it. |
|
So can you tell us specifically what you are after at the moment? |
|
Again: Open Assistant never was about offering a public chat. It was about creating a public dataset and training models that are freely downloadable. Some people from the team financed the chat privately in the last months and some companies supported this project, and they have every right to stop doing this. I would be angry if they never published oasst2 though. Your arguments would have made sense in January, but in the meantime we have tons of open models from all over the world. There is no way anyone can turn back time and somehow make this technology disappear again. If you are looking for models that you can run locally I recommend having a look at projects like GPT4ALL or Stable LM 3B. The LLM Peer to Peer project Petals is another example how it has become impossible to regulate this technology. |
The fact that models are still being trained is something to be grateful for. If you take a look at Hugging Face, you’ll notice plenty of 3-7B models that can even be run on a phone. If you can’t run those, it’s not the creator’s problem. The lack of communication from OA “leadership” on this thread is a bit disappointing. However, the main topic of this issue, hosting the models, most likely is an issue of cost. It’s not cheap to host models 24/7, and Open Assistant was never even about the model, but the dataset. The collection of data is for public release, as evidenced with oasst-1 (yes, I’m aware we’re still waiting on oasst-2). Initially, we had Pythia, then Llama 1, then MPT and Falcon, then a flurry of other models, and then Llama 2. Now we also have Mistral (7B only). The initial training of OA was on Pythia, and it was a decently capable model at the time. Things have changed, and those models are now outdated. However, the dataset can still be used to train newer and more capable models because of advancements in base models. The Open Assistant models and the dataset will never disappear, they're on Hugging Face, and they've been downloaded by thousands. |
I could make the same question, since it's crystal clear to everyone on this thread that it isn't coming back.
Stop, as in how Elon Musk, an important part of the reason of OpenAI's existance, alarmed everyone to "stop" AI, to gain some time for themselves? I don't think those "stops" are a "right" that companies have. Specially when those companies are subject to agendas.
It is something to be grateful for, only to the people that are able to run it. If there's gobernments and political dirt in there, I'm not personally grateful that THEY get that training for free.
I thought conversational chatbots from the early 2010's weren't part of this discussion. I took that for granted. And that's like saying, if you can't live in a bunker, it's your fault if a war arises and your house is destroyed, because bunkers exist. If you don't have access to one, you deserve to not have a roof to live under, and be in mortal danger. Right...... Because I chose most people in our society don't live in bunkers, and I chose to not live in one. Or some idiocy like that. Maybe it would be more realistic to say that the problem lies on the cause of the war, not on the population that die. Or to escape nuclear weaponry you could fly away from the planet, so it's people's blame too that nuclear weaponry exists and they're not the ones that can access it or decide how it's used. Whatever...... I don't have time for this kindergarten. |
I really don't understand where you are getting this from, regardless of what you believe when it comes open assistant's lack of communication it is known that the chat was taken down due to funding issues, |
Yes, you know better. Have a nice day while you waste your time waiting for it to come back. Bye. |
|
The title of this thread no longer reflects the fundamental direction of the conversation here. So I've posted my comment here: The Biggest Problem with Open Assistant Right Now I hope the OA team understands that their actions are hurting the entire independent LLM ecosystem. My comment in the other thread explains why. |
I have one account with my email address, and another one with my Google account. Yesterday all of my chats on both accounts disappeared, and now I can't start any new chats or post any messages. I closed this issue because I forgot to search before I started it. Then after not finding any current reports about it, I opened it again.
The text was updated successfully, but these errors were encountered: