The Biggest Problem with Open Assistant Right Now

[JuliaBonita]

In this thread, somebody said:

What exactly do you expect from this project? I have the feeling that you expect them to start some kind of revolution or something.

It literally says "We believe we can create a revolution" in the first sentence of the OA homepage. (See attached screenshot in case they change it.) So, yes, me and millions of other humans on this planet have reasonably expected OA to be the foundation for some kind of meaningful "revolution" for humanity.

Additionally, Yannic's intro to OA on Youtube made it absolutely clear that we should all expect a dramatic "revolution" on many levels from OA. You can quibble over how we should define that "revolution", but Yannic and the OA team were very clear about the specific ways that OA was "revolutionizing" LLM service development. So there can be no legitimate doubt that Yannic and the OA team have explicitly set revolutionary expectations.

Obviously the models are available to download, which is some consolation for the tiny number of humans with the resources to actually use them, but it is not correct to say that OA is/was only intended to build "a public dataset" (as one person said in the other thread). Nobody who has actually read the website and watched all of Yannic's videos about OA could reasonably come to that conclusion. OA has always been presented as a platform to offset BigTech's dominance in all areas of LLM service development, including chat and all other popular use cases.

Those are the facts. Now, why is OA's terrible communication a big problem? Because they have now destroyed a very large portion of the faith, enthusiasm and energy that humans all over the world had in the concept of truly "open" AI--AI that is not dominated by giant monopolies and censored by corrupt governments and the special interest groups that control them. That faith, enthusiasm and energy took many months to build.

Now all that faith, enthusiasm and energy is almost completely gone. And the longer the OA team hides in the shadows, creating massive self-induced FUD and speculation about their intentions (Are they scammers like SBF? Are they bought by BigTech? Are they controlled by self-serving billionaires? Are they [fill-in-the-blank]?), the more damage they're doing to the broader global ecosystem of independent LLM development. This is because a high-profile collapse of OA with no meaningful explanations will cause everybody who has time and resources to support this ecosystem to hesitate and be much less inspired and committed to donating their time and resources. Collectively, that's a huge loss for the ecosystem.

And it didn't have to be this way. No matter what funding problems a project has, true leaders communicate. They force themselves to be honest with their constituents. They communicate the specific steps they are taking to either resolve the problems that affect their constituents and/or communicate clearly to prepare their constituents to transition to other projects/initiatives.

Thus, the biggest problem with OA right now is their deficit of leadership. I used to love watching Yannic's videos, but after the way he and the OA team have destroyed the faith, enthusiasm and energy of this project and damaged trust in the broader ecosystem, I can't even look at his face without feeling disappointed and betrayed. That's the fundamental problem.

That's why the fundamental problem here is not a lack of funding. Funding and support naturally flow to leaders that have the experience and communication skills to inspire confidence based on the potential value/impact of their projects, rational and realistic roadmaps, and honest periodic progress updates. The OA team could have handled this situation 100 other ways that would have been more productive than this, no matter what excuses they eventually release to the public.

Yannic: You're one of the few humans on Earth that I could respect and admire as a leader of a project like this. So I'm truly saddened that you have chosen to handle this situation so incompetently. I hope you and the rest of the OA team can somehow recover from this debacle, but the damage you've already done truly breaks my heart.

At this point, the only truly open LLM project that seems like it is relatively immune to sudden collapse and has a technically and financially sustainable long-term future is Petals and any other LLM project that is building a fully distributed data processing architecture. This is the only way to avoid depending on for-profit companies and large investors that can withdraw their support at any moment for a thousand self-serving reasons.

If the OA team wants to resurrect what is left of the OA community, I hope they make a strong pivot to restructuring their code and business model as a distributed architecture and focus on aspects of LLM development that complement what Petals (and probably others) are already building.

NB: OA homepage today:

[alramalho]

Beatifully articulated. I'm with you.
I'm a contributor in the OA discord channel and will try to get an overview of what's happening and push for reorganization. Hopefully this can be reborn 🦅🔥

[andrewm4894]

business model

There is no "business model", the reality is this is an open source project and stuff like this happens. Agree there has been lack of communication recently (i have been trying to get some updates in the discord - but in a more chill and fair way given how open source projects actually work) and we do have a great community here and lots of great work done.

@JuliaBonita i agree with a lot of what you say (in terms of the community deserving better communication and being a real shame if project was to die) but also not really a big fan of your hyperbolic tone - the reality is probably much more boring, funding stuff like this is hard and expensive, people are really really busy doing other stuff (like their actual jobs and life etc), we are all basically volunteers here.

People posting big sort of "take down" issues like this (your first actual issue in the gh repo i tihnk) and using word's and language in past as some sort of "evidence" (while completely missing the whole spirit of it all) is not great way to go about it - how do you think this makes people feel who actually put a lot of time and effort into all of this. It's generally a much better approach to give as much "benefit of the doubt" as possible. In distributed projects like this stuff like this can just happen through no actual mistakes or bad intentions of any of the actors.

I hate posting anything remotely argumentative or thorny on the internet but just felt the need to try push back a little in terms of how your issue reads as I think it's a bit unfair to the actual reality or at least a little too combative for my taste and not really something I've seen before in the community (but I totally accept people can get frustrated and that's understandable too), while agreeing communication has not been great here.

That said there was announcement on the discord here of 28th September - are you in the discord? A lot of project's now tend to have more context and info in the discord so its not like there has been zero communications - a msg in the announcements channel of the discord from @yk saying

Hi everyone, we've disabled chat on the website for now (it was financed privately). We'll give you some more updates soon.
The models remain available to download / self-host / etc.

Is totally fine by me as someone who has helped out on the project. Ideally we would have had another update between then and now but I'm not going to put anyone on blast or anything - its chill - I'll ask a few more times in the discord in a nice gentle way and I'm sure there will be an update soon.

[fireYtail]

I completely agree with the OP, except on one particular point. People need to be realistic and accept that all software that you cannot run locally (you depend on servers with enough GPU power to run them), are subject to other people with enough money and power to claim it all for themselves. Because saying a beautiful speech about open AI software is very nice and all, but the moment you depend on third parties to run it - third parties which the end-user cannot control and which can be corrupted into whatever interest some people have - then in that moment there is no guarantee that it will always be available for everyone to access and use. And saying that those third parties cannot be corrupted by any means is a way of magical thinking. So, if I depend on having access myself (locally) to a 24 GB+ VRAM GPU, I'm pretty much screwed, no matter what intentions and promises the person or people behind any given software project, AI related or not, have or appear to have. Accept reality as it is, or live your lie and suffer it. The choice is in yourself, and I have made a clear decission. It's up to you now.

Edit: I have 11 GB VRAM, which is MUCH beyond the average user out there. So unless it can fit there, making some statement such as available for everyone to use is just utopian thoughts.

[JuliaBonita]

Dear @andrewm4894. I can see you're a thoughtful human. I appreciate your comments. Your comments inspired a few thoughts:

There is no "business model"

Are you sure about that? If an organization has a balance sheet and financial statements and needs to balance expenses and income, then it has a "business model". Saying "business model" is just a common shorthand for all the resource inputs/outputs associated with managing a financially sustainable company/project.

Given that OA is legally controlled by LAION, you can be sure they have a balance sheet, financial statements, accountants, and all the financial mechanics of every for-profit or nonprofit company with a "business model".

the reality is this is an open source project and stuff like this happens.

Right, which is why nearly all open source projects fail. The ones that succeed don't passively let stuff happen. That's the whole point of effective leadership, which is essential to ensure that stuff doesn't destroy projects and communities.

a more chill and fair way given how open source projects actually work

The implication here is that you know better how to manage open source projects; and "chill" management is how to manage every project. Perhaps that's your personal preference because you dislike confrontation, but respectful confrontation is often the only way that humans and organizations can improve and grow. Chill management might feel more comfortable for you personally, but it also tends to lead to stagnant organizations and missed opportunities for growth.

not really a big fan of your hyperbolic tone

Before posting, I read many posts across several different platforms to collect all the publicly available facts and community sentiment. My post distilled all the essential facts and concerns that I've already seen. If you're filtering my words through a chill lens, then I can understand how my words might be misinterpreted as hyperbolic. But I don't think that interpretation is consistent with the broader community sentiment right now, which is what determines the fate and funding of every project.

much better approach to give as much "benefit of the doubt" as possible.

Totally agree. That's why I didn't make any unsubstantiated accusations. Respectful confrontation in a spirit of encouraging improvement and growth is not mutually exclusive to giving the benefit of the doubt.

are you in the discord?

Yes, but probably less than 1% of all the people that care about OA are on Discord for various reasons. The OA team and every substantial project team knows this. That's why big announcements from large, high-impact projects are also typically posted to the homepage of project websites, not buried in obscure Discord threads.

That brings us back to the problem of leadership, which is the only claim I've been making. "Leadership" is not an amorphous, undefinable thing. It's what makes or breaks every human organization or project. No matter how much I love decentralized tech, the real-world reality is all projects will fail without effective leadership. And effective leadership can be defined very simply:

Just communicate clearly, consistently and persuasively. Tell us what problems you're facing and inspire us to rally behind you. Give us something realistic and inspiring to believe in; then the resources will naturally flow to support you. If you can't do that, then don't be surprised if your project fails due to lack of support, while other projects emerge with stronger leadership who build their success on top of your ashes.

Not hyperbolic. Just the objective reality of the real world beyond the source code.

[andreaskoepf]

@JuliaBonita I understand that you and a lot of people are unhappy about the ending of open-assistant. Our core problem was simply that all founding team members had other obligations (jobs, companies, other projects) + there was lack of qualified volunteers who wanted to contribute or continue managing the project. If you are interested in actively managing and "re-vitalizing" Open-Assistant please contact me via discord.

[aggregate1166877]

For those who haven't seen the relevant video, here's a TL;DR:

• Open-Assistant, in its current state, is meant to serve as a template for other AIs to be built upon.
• Its training data, unlike other AI datasets, does not contain copyrighted content. Every piece of training information was submitted with 100% consent, making it the most ethical dataset in existence. If you build a model with LAION's OA dataset, no one can accuse you of "stealing" from others.
• It clearly documents how it works. When OA started, the community mostly speculated about the inner workings of larger AI models. Now, we have a working examples not only from OA, but also other academic sources.
• While the project could have continued, a large chunk of contributors, after already doing significant amounts of work, lost interest and left the project. Thus, it makes sense to stop OA here as a working template. The OA project achieved its core goals and can now be the seed for the next generation of ethical AI.