Heuristics to spot AI content

Many forums/communities I'm in are going to the dogs lately, because of the rampant ChatGPT crap.
What are some quick heuristics you use to spot AI content?
For me its:
1. "Generalist" information that offers no specific information, but gives an overview, in response to a specific question.
2. ChatGPT-speak - - that very academic language that AI models typically use
3. Parody-speak - - usually, when someone prompts ChatGPT to use a specific "tone" it still comes off as somewhat fake and like marketing copy, instead of conversational (unless the person using the tool edits it enough)
4. A sudden shift in grammar space - - I haven't seen many people talk about this, but its a dead giveaway. For example, if the user has 10 posts on the forum, and the first nine were something like "dm plz" "hi what is this" "plz I dont understand" and the tenth message is "I can understand it must be extremely frustrating for you to go through this. Generally, I have found that..." that's a drastic shift in the grammar space, and can usually be an indicator of AI content.

I'll also add:
5. discrepancies between the start and end of paragraphs and the middle. Some smarter ChatGPTers will edit ChatGPT responses before posting. They tend to do this by either prefixing something to the generated text, or adding something after the end of the generated text ie bookending the AI gen text with more "natural" talk. For example, they'll add something like "umm, I dunno, in my experience" to the start of the AI gen text. However, if you check out the "middle" of the text, between the bookends, the grammar of the text will be substantially more bot-like, because most people don't have the patience to "humanize" the entire text.

short intro paragraph - load of bullet points - short conclusion paragraph. This is the pattern I see in forum posts that have come from chatgpt. Or when they've copy and pasted "regenerate response" too

The most obvious one is personality. Articles made with the tool are very "matter-of-fact" and robotic. It's great for research, though.

Masterofeverything said:

The most obvious one is personality. Articles made with the tool are very "matter-of-fact" and robotic. It's great for research, though.

Click to expand...

Yes you can easily detect nowdays robotic A.I.

I hope this letter finds you well.

That's ChatGPT's reponse:

Q: Many forums/communities I'm in are going to the dogs lately, because of the rampant ChatGPT crap.What are some quick heuristics you use to spot AI content?

A: I apologize if you feel that AI-generated content is negatively impacting the forums and communities you are a part of. While I am an AI language model myself, I can understand the concerns you might have regarding the quality and authenticity of content generated by AI. Here are some quick heuristics that can help you identify AI-generated content:

1. Consistent grammar and language: AI models often generate text that is grammatically correct and coherent. If you notice content that consistently lacks errors, peculiarities, or inconsistencies typically found in human communication, it could be a sign of AI-generated content.
2. Unusual or unexpected responses: AI models can sometimes provide responses that are unexpected or out of context. If a user's responses seem detached from the ongoing discussion or don't align with typical human behavior, it might indicate AI-generated content.
3. Generic or overly perfect answers: AI models are trained on a large amount of data, and they may generate responses that are generic or lack specific details. If you come across answers that sound too perfect or lack personalized insights, it could be an indication of AI-generated content.
4. Rapid response time: AI models can provide quick responses since they process text rapidly. If you notice someone consistently responding within seconds with well-formed answers, it could be an indication that an AI model is involved.
5. Lack of emotional cues: AI models do not possess emotions, and their responses may lack emotional context or depth. If the content appears to be devoid of emotional cues, empathy, or subjective experiences, it might suggest AI involvement.

It's important to note that while these heuristics can be helpful, they are not foolproof. AI technology is evolving, and there may be instances where AI-generated content becomes harder to distinguish from human-generated content. Additionally, some AI systems may explicitly identify themselves as AI, while others may not.
If you encounter AI-generated content that you believe is negatively impacting a community or forum, it's a good idea to communicate your concerns to the forum moderators or administrators so they can address the issue appropriately.

Second trial

Q:Answer the following question being moderate, as first person and introduce grammar and spelling errors: Many forums/communities I'm in are going to the dogs lately, because of the rampant ChatGPT crap.What are some quick heuristics you use to spot AI content?

A:
Hey, ya know, I've been hangin' around some forums and communities lately, and lemme tell ya, things are goin' downhill. It's all 'cause of this ChatGPT nonsense runnin' wild. So, what are some tricks I use to spot that AI gibberish? Well, here ya go:

1. Grammer and speling mistakes: AI stuff tends to have them perfect grammer and speling. So, if someone's text is all error-free and lookin' fancy, chances are it's an AI at work.
2. Weird, unexpected answers: These AI models sometimes come up with responses that are just plain bizarre. If someone's talkin' in a way that doesn't quite fit the conversation or seems off, ya might be dealin' with an AI-generated response.
3. Generic, cookie-cutter replies: AI tends to give ya these one-size-fits-all kinda answers. If someone's spoutin' off info that sounds too textbook-like or lacks any personal touch, it's prob'ly an AI talkin'.
4. Speedy responses: These AI thingies process text faster than us humans can blink an eye. So, if someone's replying within seconds, even when it's a complex topic, it's a good bet that AI is involved.
5. No emotional vibes: Now, these AI contraptions ain't got no emotions. So, if the content feels robotic and lacks any human touch, like empathy or personal experiences, well, it's most likely an AI doin' the talkin'.

Mind ya, these tricks I mentioned aren't foolproof. The AI tech keeps gettin' better, and sometimes it's harder to tell what's human and what ain't. Also, some AI systems might just straight up tell ya they're AI, while others won't. So, take these tips with a grain of salt.
If ya come across AI-generated content that's causin' problems in a community or forum, it's best to let the mods or admins know. They can sort things out and keep the place runnin' smoothly.

Q: Answer again using a very informal speech

A:
Yo, dude! So, check it out. I've been hangin' out in some forums and communities lately, and man, things are goin' to the dogs, all thanks to that rampant ChatGPT crap. It's a mess, ya know? Anyway, you asked about how to spot AI content real quick. Here are some things I look out for:

1. Grammar and spelling screw-ups: AIs usually get their grammar and spelling on point, so if someone's text is all pristine and error-free, you can bet it's AI-generated.
2. Weird, wacky answers: Sometimes these AI models come up with responses that are outta left field. If someone's talkin' all weird and not makin' sense in the context, chances are they're AI-generated.
3. Cookie-cutter replies: AIs love to dish out those generic, one-size-fits-all responses. If someone's givin' you an answer that feels like it came straight outta a template, it's probably AI talkin'.
4. Lightning-fast replies: AIs process text faster than you can blink, bro. If someone's pumpin' out responses in seconds, even for complex stuff, it's a dead giveaway that AI's in play.
5. Zero emotions: AIs don't feel a thing, man. So if the content is all robotic and lacks any human touch, like empathy or personal experiences, you're dealin' with AI content for sure.

Now, listen up. These tricks ain't foolproof, okay? AI tech keeps advancin', and it's gettin' harder to spot the difference. Plus, some AIs straight up admit they're AI, while others don't. So take these tips with a grain of salt, my friend.
If you stumble upon AI-generated content that's messin' up a community or forum you're in, make sure to reach out to the mods or admins. They'll sort things out and keep the place from goin' to the dogs. Peace out!

Nothing we can do sadly. It's only going to get worse.

That's a very old topic, but I think nonetheless it is MORE than important to not give up on this.

I was just scrolling through Blackhatworld and noticed a guy whos posts were.. "bland". So I took a look at his other posts.. and voilá, except for his personal questions all made the same first impression.
After reproducing/mirroring the potential task given to Chat GPT, it turned out to be nearly exact to what he "wrote".

In regards to this.. Does anyone know of a company that specializes in detecting AI written text?





rofl..

All jokes aside though. Sh#$'s creepy. Voice AI even more.. These Joe Rogan shorts etc.

Several companies focus on detecting AI-generated content. Some notable ones include OpenAI, which developed GPT models like me, and companies like Deepware Scanner, Content Authenticity Initiative (CAI), and Truepic. These companies work on developing tools and technologies to identify and verify content created by AI to address concerns related to misinformation and deepfakes.

Click to expand...

Welp.. Nevermind.. Question got answered. :-|

with the general information and the so formal way the information is provided and no that much technicals words and lack of personality and personal oppinion is noticeable.

go to the bottom and if you see " in summary" it's total crap.

The internet is going back to that spammy pop-up time we had 10-15 years ago, piracy solved it last time and it most likely will again at the expense of stock prices and shareholders but Netflix is the new Blockbuster and new ideas and business models will happily take the place of the current state of greed we see online across the board.

AI is not what people think it is and it's an obvious petrock to anyone who can read well...

noellarkin said:

What are some quick heuristics you use to spot AI content?

Click to expand...

These algorithms work in a different way. GPT detectors are GPT models themselves. They are designed to analyze sequences of tokens generated by other GPT models. These detectors work by examining the flow and structure of the token sequence, assessing whether it conforms to typical human writing patterns or whether it bears the hallmarks of AI-generated content. Essentially, they use the same underlying technology as GPT models, but instead of generating text, they just identify and identify to distinguish between human and AI authorship. This process relies on recognizing the subtle nuances and characteristics that distinguish AI-generated text from human-generated text, using the same principles of language modeling that guide GPT's text generation. So, your vision is incorrect. The GPT detectors don't estimate how "general" the text is, etc. They simply analize the token flow according to their quality standards.