I don’t worry about that. There are other markers of AI that are much more reliable:
Extreme verbosity.
Low S/N - i.e. lots of words to say not much at all.
Perfect grammar.
If you drill down into the subject, often completely incorrect - but you don’t know without having to read a whole bunch of tedious text.
And here’s how you recognize AI:
High-schoolers turning a paper on a subject they know nothing about often fluff up their paper - at least when students still wrote their papers themselves - and hit 1. and 2., but rarely 3.
Good writers always hit 3. They can be terse or verbose, and they may or may not hit 1., but never 2. or 4.
Internet writers don’t write like journalists. Only journalists writing for a journal that happens to also publish on the internet write like journalists. Internet writers don’t quite hit 3, knowledgeable ones don’t hit 2., and almost none of them ever hit 1. Or said another way, when you read something about Linux networking that looks like an Atlantic op-ed, it’s AI.
Only AI hits 1., 2. and 3. AI almost always writes in a tone and form that doesn’t befit the venue.
As for 4., if you want an example of this, try to search “NFC unlock” on DDG or Bing (same AI-laden Microsoft trash search engine): you will find scores of perfectly-written articles that explain in painful details how you should buy NFC tags (they don’t say which), program them (they don’t say how), then present the tags to your device (they don’t say what devices) to program them to unlock upon presenting the tags.
If you know anything about NFC, you know this is all shades of wrong. But amazingly, each article on the subject is many pages long, perfectly written, and there are countless such articles.
I find it interesting that AI writes very high level, which is a trait that seems to be valued by upper management (VP, CEO level).
I think at some point someone will figure out that AI would sound more natural if it didn’t write so perfectly, and they’ll try to make AI sound hip and casual, and I’m almost certain AI will still fail to be convincingly casual.
I haven’t really tried but I think if you were trying to use LLM generated text deceptively the model could probably do exactly as you describe already so long as you prompt it to do so clearly enough.
You can write in the prompt to the AI to write in hip and casual tone, but you will get something that will give you “How Do You Do, Fellow Kids?” meme vibes.
I don’t worry about that. There are other markers of AI that are much more reliable:
And here’s how you recognize AI:
High-schoolers turning a paper on a subject they know nothing about often fluff up their paper - at least when students still wrote their papers themselves - and hit 1. and 2., but rarely 3.
Good writers always hit 3. They can be terse or verbose, and they may or may not hit 1., but never 2. or 4.
Internet writers don’t write like journalists. Only journalists writing for a journal that happens to also publish on the internet write like journalists. Internet writers don’t quite hit 3, knowledgeable ones don’t hit 2., and almost none of them ever hit 1. Or said another way, when you read something about Linux networking that looks like an Atlantic op-ed, it’s AI.
Only AI hits 1., 2. and 3. AI almost always writes in a tone and form that doesn’t befit the venue.
As for 4., if you want an example of this, try to search “NFC unlock” on DDG or Bing (same AI-laden Microsoft trash search engine): you will find scores of perfectly-written articles that explain in painful details how you should buy NFC tags (they don’t say which), program them (they don’t say how), then present the tags to your device (they don’t say what devices) to program them to unlock upon presenting the tags.
If you know anything about NFC, you know this is all shades of wrong. But amazingly, each article on the subject is many pages long, perfectly written, and there are countless such articles.
Wow you explained AIs tone and writing very well.
I find it interesting that AI writes very high level, which is a trait that seems to be valued by upper management (VP, CEO level).
I think at some point someone will figure out that AI would sound more natural if it didn’t write so perfectly, and they’ll try to make AI sound hip and casual, and I’m almost certain AI will still fail to be convincingly casual.
I haven’t really tried but I think if you were trying to use LLM generated text deceptively the model could probably do exactly as you describe already so long as you prompt it to do so clearly enough.
To be honest I thought AI was already used to flood popular message boards with user content trying to sound natural to show user engagement.
Yeah for sure, but there’s always something off that screams fake.
That’s why I prefer lemmy! Real users.
First it was trolls, then it was trolls using AI. I understand why many users are abandoning social networks.
Trolls arguing AI bots would be a whole new level lol
I think, part of the “problem” is that most of their training data is formal language. They’ll have much less data from private chats and such.
Meta has their own AI Llama. I’d imagine meta would have private chat data.
You can write in the prompt to the AI to write in hip and casual tone, but you will get something that will give you “How Do You Do, Fellow Kids?” meme vibes.
I thoroughly enjoyed what you did with the formatting there.