We have a couple of folks who are taking on AI work in a pretty individualized/haphazard way from what I can tell. While this is an important first step, this approach is insufficient to meet the challenges ahead of us imho.
It seems like we should organize amongst ourselves both as MLs, and also as an antidote to the inherently antisocial nature of the tech itself
Similarly, there’s currently no ML AI tech except that shit chatgpt wrappers by the WSWS.
I’m not particularly prescriptive about the form it takes, but just wanted to start the conversation
Let’s build collectively rather than individually
You must log in or # to comment.
Brillant ending. I too think that we need to “organize ai” and have stuff like red creators network on youtube readily at hand for everyone, same as a most complete set of current and former standard work of marxist leninist literature so we have a simple q&a style that can literally answer every theory fact there is and probably even show similarities between works or holes that need to be closed, etc. This would of course have the potential effect that one can copy paste the an attack from liberals the same way they do themselves and immediately see the flaws in their thinking. One could train on a “liberal from hell” who has just enough knowledge to make you go insane with their arrogance and so on.
That said, we have multiple comrades working with and on AI to make it work for us. If they could state their current status it would probably make sense.
Couldn’t agree more comrade
And it’s the previous duplicative efforts to gather all the Marxist literature that inspired this post. GitHub/hugging face are scattered with various attempts to build this text base, not to mention individuals here doing the same thing by scraping MIA/Prolewiki
When we do not collaborate, its actually harmful imho to
- the planet, unless you’re running all of this locally, you’re contributing to the pollution and water consumption crisis. And even if you’re are, you need to be on a clean grid to actually mitigate the consequences. By pooling resources, we can pursue training and inference to bring this tooling to the community with a known carbon and water footprint of nearly zero
- the proletariat, the Q&A use case you made above is a great example of something that seems simple, but is quite hard and expensive to do in practice (we need to build up a dataset of exactly this style of question at a high rigor/academic level, and in a way where we don’t “leak liberalism” into the dataset or model. Nothing we make will be useful unless it’s trained on that data, so it makes zero sense for us to compete in creating that dataset rather than collaborate
- the lay public and future generations, forgive the grandiose claim, but when communists have no material alternative to the techno fascist colonization of the future, we abandon those on the sidelines to “choose” this dystopian future because they have no other options
To directly answer your question, I’m personally finishing a master’s degree this year, so I’m looking to root my knowledge in the community rather than the market. I’ve got local models up and running, and my next step is to build up that exact text archive you mentioned above
Sounds great. Let me know if you make efforts to organize this. I’ll gladly join a group and take on tasks.
Perfect and I’m looking forward to building with you comrade. I’ll send a when2meet tomorrow to interested parties so we can determine a good initial meeting time.
The list so far is,
If anyone else is interested or I missed anyone, as their name in this thread so they’re included
yeah we could chat on matrix or something, or even just a lemmygrad thread could work, that way we’ll have an easy reference and meeting notes :)
Starting a discussion about the direction we want to see would be very useful. Very much agree that we have to start working on this collectively, and we have to come up with a concrete vision for what we aim to build.
I’d argue that core principle should be that the models have to be open, doing what WSWS did with simply calling out to a proprietary service is not the way to go, these have to be tools that we own and control.
We should also aim to make models that can be run locally which means focusing on 32bln or less params. Figuring out how to improve the quality of output for these models would be really important.
The approach that petals is doing also needs to be explored, where they use torrent style distributed framework for training models. This avoids the need for big data centres for training new models.
And @haui@lemmygrad.ml already did a great job covering the content creation aspect.
Thank you. I’m glad to help. Let me know what you need. o7
It might be good to figure out who’s all interested in this idea, will likely have some ideas here too. :) Maybe all of us who are interested in this could get together and talk through the ways we envision models being used, and how to facilitate that.
There are basically three main aspects in my mind. First is education on why these tools are useful in general, why they shouldn’t be shunned, and how to apply them effectively. This is particularly important for people who aren’t developers themselves. Having good guides, and explanations. @CriticalResist8@lemmygrad.ml has already been doing some great work on that front with the crush community.
Then there’s the aspect of how to apply these tools politically for effective agitation, making memes, writing essays, etc. This happens ad hoc right now, but it would be good to start building consistent Marxist messaging in general.
And finally, there’s the development of these tools which falls to technical people with the expertise to make that happen.
I would think a good approach to start would be.
Create a list of questions that need to be clarified before solving problems. Create a list of issues/tasks and what is required to solve each problem.
If no one knows the answer right away, people can analyze and write the answers according to their opinion.
