FB has been junk for more than a decade now, AI or no.

I check mine every few weeks because I'm a sports announcer and it's one way people get in contact with me, but it's clear that FB designs its feed to piss me off and try to keep me doomscrolling, and I'm not a fan of having my day derailed.

If you could reliably detect "AI" using an "AI" you could also use an "AI" to make posts that the other "AI" couldn't detect.

The context is bad though.

The post I'm referencing is removed, but there was a tiny “from gemini” footnote in the bottom that most upvoters clearly missed, and the whole thing is presented like a quote from a news article and taken as fact by OP in their own commentary.

And the larger point I’m making is this pour soul had
no idea Gemini is basically an improv actor compelled to continue whatever it writes, not a research agent.

My sister, ridiculously smart and more put together, didn’t either. She just searched for factual stuff from the Gemini app and assumed it’s directly searching the internet.

AI is a good thinker, analyzer, spitballer, initial source and stuff yes, but it’s being marketed like an oracle and that is going to screw the world up.

Sure, but then the generator AI is no longer optimised to generate whatever you wanted initially, but to generate text that fools the detector network, thus making the original generator worse at its intended job.

I deleted facebook in like 2010 or so, because i hardly ever used it anyway, it wasn't really bad back then, just not for me. 6 or so years later a friend of mine wanted to show me something on fb, but couldn't find it, so he was just scrolling, i was blown away how bad it was, just ads and auto played videos and absolute garbage. And from what i understand, it just got worse and worse. Everyone i know now that uses facebook is for the market place.

It's such a cesspit.

I'm glad we have the Fediverse.

I did not know that. There’s a bunch of news articles going around claiming that even the creators of the models don’t understand them and that they are some sort of unfathomable magic black box. I assumed you were propagating that myth, but I was clearly mistaken.

That’s an extremely low sample size for this

I see no reason why "post right wing propaganda" and "write so you don't sound like "AI" " should be conflicting goals.

The actual argument why I don't find such results credible is that the "creator" is trained to sound like humans, so the "detector" has to be trained to find stuff that does not sound like humans. This means, both basically have to solve the same task: Decide if something sounds like a human.

To be able to find the "AI" content, the "detector" would have to be better at deciding what sounds like a human than the "creator". So for the results to have any kind of accuracy, you're already banking on the "detector" company having more processing power / better training data / more money than, say, OpenAI or google.

But also, if the "detector" was better at the job, it could be used as a better "creator" itself. Then, how would we distinguish the content it created?

If you want to visit your old friends in the dying mall. Go to feeds then friends. Should filter everything else out.

8,855 long-form Facebook posts from various users using a 3rd party. The dataset spans from 2018 to November 2024, with a minimum of 100 posts per month, each containing at least 100 words.

seems like thats a good baseline rule and that was about the total number that matched it

With apparently 3 billion active users

Only summing up 9k posts over a 6 year stretch over 100 words feels like an outreach problem. Conclusion could be drawn that bots have better reach

Not my Annie! No! Not my Annie!

each post has to be 100 words with at least 100 posts a month

how many actual users do that?

And 58.82% are likely generated by human junk then.

I have no idea because I don’t use the site

My brother gave me his Facebook credentials so I could use marketplace without bothering him all the time. He's been a liberal left-winger all his life but for the past few years he's taken to ranting about how awful Democrats are ("Genocide Joe" etc.) while mocking people who believe that there's a connection between Trump and Putin. Sure enough, his Facebook is filled with posts about how awful Democrats are and how there's no connection between Trump and Putin - like, that's literally all that's on there. I've tried to get him to see that his worldview is entirely created by Facebook but he just won't accept it.

In my mind, this is really what sets social media apart from past mechanisms of social control. In the days of mass media, the propaganda was necessarily a one-size-fits-all sort of thing. Now, the pipeline of bullshit can be custom-tailored for each individual. So my brother, who would never support Trump and the Republicans, can nevertheless be fed a line of bullshit that he will accept and help Trump by not voting (he actually voted Green).

I don't use the site either but 100 words is a lot for a facebook post

That laptop lol.

I'm not necessarily saying they're conflicting goals, merely that they're not the same goal.

The incentive for the generator becomes "generate propaganda that doesn't have the language chatacteristics of typical LLMs", so the incentive is split between those goals. As a simplified example, if the additional incentive were "include the word bamboo in every response", I think we would both agree that it would do a worse job at its original goal, since the constraint means that outputs that would have been optimal previously are now considered poor responses.

Meanwhile, the detector network has a far simpler task - given some input string, give back a value representing the confidence it was output by a system rather than a person.

I think it's also worth considering that LLMs don't "think" in the same way people do - where people construct an abstract thought, then find the best combinations of words to express that thought, an LLM generates words that are likely to follow the preceding ones (including prompts). This does leave some space for detecting these different approaches better than at random, even though it's impossible to do so reliably.

But I guess really the important thing is that people running these bots don't really care if it's possible to find that the content is likely generated, just so long as it's not so obvious that the content gets removed. This means they're not really incentivised to spend money training models to avoid detection.