Thanks I love Juche can’t wait

Join prolewiki and upload books from Anna’s archive to our text library, you can even make categories for reading lists etc.

no you see, we shouldn’t be moving at all, that way we won’t notice our chains and if we don’t notice them it means they don’t exist.

he’s been getting more and more terminally online

Yes, but they become fragmented and divided. This is why deplatforming works. The people that used to frequent these places still exist after the ban and still browse reddit, but instead of congregating in one unified place they might browse reddit less and turn to closed discords, or join a number of smaller, new subreddits instead.

On censorship:

Communists are not scared of censorship. To capitulate to the censors is to admit from the start that we don’t believe in our own theory. It’s opportunistic, because it leads one to the depths of liberalism. After all if one is scared of being censored for being a communist, then the only solution is to toe the line and you might as well become a democrat if you’re scared of saying anything that could get you banned from any big platform.

The solution is to take control of our own tools, and build dual power. And I don’t mean this online specifically, I mean in every aspect of life. Many parties today make the mistake of capitulating to censorship before they are even censored, because they’re scared. Then they become reformist parties, indistinguishable from liberals, diluting theory to become socially acceptable and palatable as the material conditions continue dwindling around us. Anything to get on TV.

Places such as Lemmygrad allow us to congregate, talk and have a community. But more than that, they also allow us to share knowledge and theory and learn from each other. The solution is to have both self-owned platforms and continue the struggle outside of our bubbles too. It’s not all one or the other, it’s both.

Lemmygrad can’t be banned - first of all it’s all federated. Even if lemmygrad stopped existing, every post that’s federated on the dozens of instance we are linked with would still be up. You can’t take down the federated web unless you take down every single federated website. Moreover, lemmygrad is self-hosted. Being self-hosted, the website can also be brought back very quickly. This is the case for any self-hosted website if you run daily backups like you should be doing anyway.

he died as he lived, spouting racist bullshit

You can sign up on any matrix instance, the ‘flagship’ one is matrix.org but there’s others. Save your recovery keys when creating your account, in your notes app or .txt file on your phone or desktop computer. Don’t give them to anyone, and don’t delete them either. Save in multiple places (drives and devices). It’s just a text file, it doesn’t take a lot of space, but if you lose them you won’t be able to log back into your account.

Then you have to connect from a client. There’s a ton of them because matrix is open source. There’s clients on phones, web tabs, or desktop apps. Personally I use Cinny, it’s similar(ish) to Discord and so you probably won’t get too lost on it. Cinny to my knowledge only exists as a web app, app.cinny.in. Log in with your account and provide your decryption (recovery) keys when prompted or in the settings so that you can read encrypted messages. If you didn’t save your recovery keys and your account is new you can always just recreate a new one.

Third step is to connect to the genzedong server. The address should be in this thread: https://lemmygrad.ml/post/1294067.

But to join a server on matrix, you don’t provide a link but an address to a room (channel). To do that on Cinny, click the + button, ‘add space’ -> ‘join with address’.

And then send the address for the rules channel and the other two channels as provided in the thread above (the ones in gray that start with a #). I won’t paste them here just in case they change in the future, it’s better to refer to the source material.

You also have to manually join the vetting channels and answer the questions and after that if you’re approved by the admins, they’ll invite you to a space [I think that’s how they do it], which is kinda like a server on discord (more technically a space on matrix is a collection of channels).

From that space, you will then have to join the channels you want to see manually. On Cinny once you’re in the space you need to click on the ‘lobby’ in the second sidebar to see all the channels.

I tried your proofread method with another file and I think there’s definitely some merits, to make sure that specific terms get translated the same way each time and improving consistency. I just asked deepseek to do a second pass and look for consistency, typos, errors, etc. Didn’t seem to have a lot to correct though.

For books I break it up, but deepseek seems to be able to handle a huge amount of tokens. If it can’t handle it anymore (if the convo gets too long), it will return a server error so I just copy my initial prompt and a long portion of text and start over in a new chat.

Just to make sure, I tell it in my initial ‘framing’ prompt that I’m going to be sending excerpts sequentially and that it should only return the translation and nothing else.

For mistakes etc you could probably ask it to do a second pass. You might even want to try a new, fresh chat so that it doesn’t know what the original was. That’s a good idea that I hadn’t thought about!

I was just thinking that it would be easy to remove the keys for reduced tokens but treating it as extra context for the LLM makes a lot of sense.

And it saves on effort too if you just send it the full file x)

I definitely need to get in on agents, I’ve been wanting to bring custom changes to my apps for a while now, or even make simple scripts for myself and my workflow, and having to feed the AI the code manually has been pretty limiting.

One could also imagine bringing these custom additions to the codebase too to keep foss projects thriving, but probably in a less “this is my very specific solution” and more “you can adapt my specific solution to be a little bit more specific to you” (e.g. in keyboard shortcuts, which I love having and would love to add to some apps I use).

I think you could definitely help with that, I’ve started the collab document and can send you the link

This is exactly how I would submit anything AI to a collaborative project. On ProleWiki we’ve opened page editing to users without an account and often get low-quality edits, which we then have to take time to read through and reject. It’s on us to provide a proper document that explains how people should edit. We don’t have to face AI edits yet, but at some point honestly they will be indistinguishable (I already use AI in writing to like clarify a sentence that I’m cornering myself into and other quick stuff like that). A lot of low-quality AI contributions are probably well-meaning (or not-so-well-meaning) people who just copy and paste whatever the AI gives them without a second thought.

I think they might want to also give tips on how to prompt and what to tell the AI (and make sure that it takes into account). It still takes down time on some tasks if you wouldn’t be efficient doing them yourself. Instead of learning to code this (months), you can properly set up your AI chat to get a workable solution in maybe 30 minutes to an hour. It’s not the 10 seconds it takes to say “hey can you code this for the curl project?” “sure here’s some silly whatever code” but it cuts down a lot. Someone some time ago sent me a link to an engineer working with a Unitree humanoid robot at home, and he used AI agents to do basically all of it in just one or two hours for every new functionality he wants.

I totally get you. I think AI has the potential to get more people interested in FOSS, and contributing too. I now take the view that they will only get better at handling codebases, not worse or plateau, and it will allow everyone to start making apps or even modifying them for themselves. There’s even a lot of open-source models that can run locally on machines we already have, so their power consumption is nothing different from running a modern triple A game or rendering software (and no water wastage in the machine itself). It might not be at that level yet for collaborative projects, but imo it’s important for FOSS devs to look into it and answer the question regarding their projects and needs. I have ideas for some projects (incl. Lemmy) and would love to contribute working code to it instead of just being an ideas guy.

I was surprised by local models. I know someone who has two 4060s 16GB in their PC. They run local models with both and their power supply unit is something like 650 watts? Numbers that have been in computers for decades basically.

The overarching problem is getting that energy tbh, nothing new. Every building should be adorned with chinese solar panels but we artificially protect “our” industry (the non-existing one in Europe lol) so you can’t get chinese solar panels without huge tariffs and fees.

I can probably set up a Google doc or similar (I like cryptpad but it’s slow) and send you and memorablename the link to collaborate on it, and anyone else who wants to participate. Though in my experience it’s good to limit the authors to 3 or 4.

To add to the last part, my preferred way is to get acquainted with the library/framework if I’m gonna be using it a lot, and then complete everything else with AI. That way I still learn and know how it works under the hood so I can also guide the AI if it starts getting off topic.

It’s a teaching by example tool. I don’t necessarily read or review the code but I ask it, why do it this way? Wait, I didn’t think you would do it like that, explain?

A lot of the time documentation is severely lacking or meant for other devs. I remember getting in bootstrap years ago took me weeks. With AI I could probably get around to it in an hour.

this is an overstatement. once you learn the basics of one programming language (which does not take a full year), you can apply the knowledge to other programming languages, many of which are almost identical to one another.

I’ve tried getting into javascript at different points. My brain doesn’t like OOP for some reason. Then after that you have to learn jquery, then apparently React or Vue.js… That’s when I stopped looking lol because as much as in my job knowing web dev is useful I’m not a frontend dev either.

I could maybe get something working after 6-9 months on it, if I don’t give up. But it would be inefficient, amateurish and might not even work the way I want it to.

I’m not even talking about full apps with GUIs yet, just simple-ish scripts that do specific things.

Or I can send the process to AI and it does it in five minutes. By passing it documentation and the code base it can also stay within its bounds, and I can have it refactor the code afterwards. People say it has a junior dev level and I agree, but it may not stay that way for much longer and it’s better than my amateur level.

To say “you must learn programming it’d the only way” was true only before 2022. I would still say it’s good/necessary to know how code and computers work so you know how to scope the AI but aside from that like I said we don’t always have a programmer friend around to teach us or make our scripts for us (as much as I love them)

I was gonna say, it seems to me project heads need to declare how they want to handle AI contributions and not accept all contributions instantly. Make a document explaining how to contribute AI code, how to prompt it etc.

I mean anyone could code malicious code and commit it if they wanted to.

What are their arguments against it?