99
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 24 Jun 2023
99 points (100.0% liked)
Technology
37739 readers
481 users here now
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 2 years ago
MODERATORS
I don't know. Social media bots have been doing exactly that quite well for a long time. Turns out, you don't actually have to write a comment, you just need to find another one that talks about the same key words and copy it in.
You still get great natural language (since it is natural language) and it fools most people as well.
Political talking points aren't that varied. There are a handful of different takes on each topic and people repeat them already, so just copying them doesn't make much of a difference.
It's not the same. GPT-based bots add much more to the situation.
Current bots are easily identifiable, and can be just banned when spotted, but gpt bots can interact in a way that makes is more difficult to spot. They can be programmed to present different personalities and tastes, commenting on several places, and even chit-chatting here and there. Then, they will do their propaganda, considering the contexts, arguing and replying to counterarguments.
It's a much more complex structure, and much harder to identify. Today, gpt produces text following some patterns, but that's something that can be improved.
All we can really hope for is effective AI-driven detection methods for AI generated content. Here's hoping that AIs are good at spotting one another.
That's not a workable solution. Since Meta's algorithm was leaked, there has been such rapid advancement on the open-source side of LLMs that the tech has diverged too far to ever be detectable.
You can now spin up a custom, targeted LLM in a few hours on low-power consumer hardware. And it beats the massive incumbents within the narrower scope of the training.
Think, a Facebook comment bot, targeted specifically to sound like pro-[VIEW] comments, complete with typos and Internet slang. Or a high school essay bot, trained exclusivity on 5-paragraph essays.
The tech right now gives a bet high false positive rate, and there are also AI tools that rewrite text to avoid detection by the existing detection tools.