I already made some people mad by suggesting that I would I would make by computer run an ollama model. I suggested that they make a counter AI bot to find these accounts that don’t disclose they’re bots. What’s lemmy opinion of Ai coming into fediverse?
You must log in or # to comment.
I’m sympathetic.
But… What exactly would you use them for? Spam detection would be quite expensive, in other cases it’s basically a writing assistant for a human response.
If you’re talking about the counter AI measures I’m curious if they exist and I want to implement them in a bit that makes human like responses. But the AI I’m curious if it can the turing test
If it added value then I wouldn’t be opposed. But I don’t see what value AI could possibly add to a social network. Some specific fields, like researchers combing through large data sets, have benefitted from AI. Every other place it’s been shoehorned into has suffered for it.
If you see a problem and realize AI could address it, then that’s fantastic. If you’re coming at it from the other direction and looking for problems then you’re going to waste everyone’s time.
While I am an AI enthusiast, generative AI has two issues that make it very hard to accept here
One is definitely the fact that we all know they have been trained using our data without our informed consent, not to mention it bring a typical case where copyright only applies to big companies, it doesn’t really protect individuals.
The second one is simply that we are in a social network. Social. We use it to communicate with people, not to play games or take part in experiments. It’s like using comments to a question for statistical purposes, you have to tell people they are taking part in it.
Here we want to discuss daily life, politics and hobbies with other people, forming opinions based on what other people think, and spending time and energy to explain our positions to other people. If the other end is a machine, how is this different from an NPC from an RPG game?
So, I guess the only way to go for it is to have separate communities that specifically allow AI bots, making sure people know about it so they take part if they are willing. Ofc we can expect some instances deciding to cut ties with AI filled ones, it’s up to them to decide.
We have some cracking communities for AI images –
The moral panic where the hivemind loves-to-hate AI won’t last, I just tune it out.
I love genAI and I play with it all the time. I also use it to generate inspiration for my art. I’d never suggest releasing a model to the Fediverse.
“People are mad I want to make a spam bot”
I stopped caring what other people think. Especially when they can’t say why they’re mad
Kill it to death. With hammers.
In general, if it isn’t open source in every sense of the term, GPL license, all weights and parts of the model, and all the training data and training methods, it’s a non-starter for me.
I’m not even interested in talking about AI integration unless it passes those initial requirements.
Scraping millions of people’s data and content without their knowledge or consent is morally dubious already.
Taking that data and using it to train proprietary models with secret methodologies, locking it behind a pay wall, then forcing it back onto consumers regardless of what they want in order to artificially boost their stock price and make a handful of people disgustingly wealthy is downright demonic.
Especially because it does almost nothing to enrich our lives. In its current form, it is an anti-human technology.
Now all that being said, if you want to run it totally on your own hardware, to play with and help you with your own tasks, that’s your choice. Using in a way that you have total sovereignty over is good.
There are totally open efforts like IBM Granite. Not sure what is SOTA these days.
There are some diffusion models like that too.
Problem is there’s a performance cost, and since LLMs are so finicky and hard to run, they’re not very popular so far.
Apache opens weights is good enough for many cases. Sometimes the training stack is open too, with only data being the morally dubious closed part.
LLMs, image generators like Stable Diffusion etc, and other of what’s come lately to be called “generative AI” should have no place on the Fediverse or anywhere else.
Personally, if I see AI content I block the user that posted it. If a community is all about AI, I block the community. I want to see content from people that have actual talent or something intelligent to contribute.
