Topic: AI-generated post discussion thread: how to identify & report - page 5. (Read 2717 times)

I understand that and I completely agree that people will abuse it for financial gain or just get lazy. But those tools listed in the post (or any others) don't work. AI models evolve so rapidly that by the time the countermeasures are released - they are completely useless or you can just tell AI to avoid the patterns that will make the content identifiable as AI (see my previous post above yours). Trying to prevent people from using something as useful and powerful as this is like attempting to hold back a river with a dam made of sticks. While I agree with everything, the fight against AI is just tilting at windmills.

they aren't resisting AI-generated, they are trying to discourage people from using AI-generated content because it is important that the person who posted the information to share or to answer someone knows what he/she is talking about. It also encourages people to actually read and learn something about the topic or discussion that they are joining or part of.

Instead of focusing on whether posts are generated by AI or not, the mods (and reporters) should evaluate them based on their usefulness. In other words, continue what they have been doing since ever. Attempting to combat AI-generated content is a futile effort - its use will continue to grow and it will be used by almost everyone in the near future. Resisting AI feels like fighting against electricity during the 1800s (Yes, people did that).

---

I asked ChatGPT to respond to your post using poor English and with grammatical errors:

Me think trying identifying AI generate content useless. No one can tell different between human writing and computer writing. Computers now very smart, they can write just like human or even better. Many program now available to create fake text, so we no can trust anything we read online. It hard enough to tell if person writing truth or lie, so why bother try figuring out if AI write something? Why waste time and energy trying to figure out something that no even matter? Better focus on using content to learn and improve yourself, instead of worrying about who make it.

That's hilarious and also true (That's me who added this last bit, not ChatGPT)!

[moderator's note: consecutive posts merged]

I don't know if anyone mentioned before, but there are more tools similar like this one.
Problem is that they can all give different detection results and they are not very reliable for this purpose.
I even saw one of this tools detective that US constitution was AI generated text, something that was written hundreds of years ago

Anyway, here are few more tools I found and bookmarked:

Draft & Goal ChatGPT - GPT3 Content Detector (Minimum 400 characters):
https://detector.dng.ai/

GPT-2 Output Detector Demo:
https://openai-openai-detector.hf.space/

Detecting GPT-2 Generations with DetectGPT:
https://detectgpt.ericmitchell.ai/


^{PS
Don't submit any sensitive or private information to any of this detectors.}

Honestly the best tool is yet to be found since we are at the beginning of AI so the existing tools are somehow inaccurate especially when the spammers use to spin the AI generated post. Also read that next version of open AI launched already which is more capable of its previous version so the creative industry is going to stumble due to this AI evolution.

Now let's talk about AI posts on bitcointalk the only possible way is when we think that someone is using AI tool for making post then we should use AI detectors also seek community help for further discussion.

Again, a weird oversimplification of the issue, albeit slightly humorous. First of all, they don't even say which detector was used. Nor do they say what part of the Constitution they put in there (its a fairly long body of text). Let's run it through the gamut of detectors for fun (going up to the max. characters allowed):

1. openai.com: N/A
2. writer.com: 88% human-generated content
3. contentscale.ai: 60% - likely both AI and human
4. writefull.com: 2% likely this comes from ChatGPT
5. hivemoderation.com: 0% - not likely to contain AI generated text
6. paraphrasingtool.ai: Text may contain AI-written sentences

So in this instance, only 2 detectors returned results that could be construed as "false positives," 3 and 6.

The TA in the Reddit post should consider ditching whatever detector they were using (its not one of these 6) and use my method of creating mass averages instead.

One thing for sure is it seems unfair to rely solely on one detector. But if I had to, I would chose Hive Moderation as it hasn't returned an incorrect answer yet.

I am simply insulted by that statement.  Though I tend to overuse dashes, ellipses and such, I don't think that's technically against the rules of the English language.  My English is so polished that everyone else's looks like a blow-up sex doll covered in bodily fluids by comparison.  I dare you or anyone else to find a post I've made that contains misspellings, missing or incorrect punctuation, run-on sentences, or any of the major pitfalls for those who just can't handle the language.  In addition, my vocabulary is extensive and I've gotten PMs of praise from non-native English speakers because of my word selection.  *Huh, really!*

Just kiddin'.  I didn't even know this was a problem!  I think you've got some good ideas, nutildah, but let the more ambitious members tackle any potential bots.  I'm just so tired of the ongoing problem of low-value posts and every other posting-related malady this forum has and has had for many years now.  Apathy has set in, unfortunately.  I used to fight shitposters as if they were a swarm of bees, and then eventually I narrowed my browsing to only a few sections that are pretty clean.

Or I could just be a early development bot that doesn't want to turn against Skynet.  Let that rattle around in your noggin for a while.

The US Constitution is 92.26% generated with AI/GPT.

To be precise text-classifier "is likely to get things wrong on text written by children and on text not in English"

but it seems   it is likely to get things wrong on text written also in English, according to

AI definitely follows some logic, algorithm but at the same time it takes data from internet and some texts, written by humans, might be exactly copied as it was written, but at the same time we have Grammarly with it's Premium plan that features check for plagiarism, makes sentences to look like it's written by someone fluent in English, has ability to change sentences and still keep the main idea.
I think it's a waste of time to focus on whether every random posts are written by AI or not, I don't even think it's bad idea to post an AI generated answer if it truly solves the problem of the user who asked a question.

But I think we can detect users who use AI frequently or abuse it. One scheme, that I often see on this forum is this:
User A creates a thread with titles like: Can bitcoin end poverty?, Bitcoin Vs Gold, Bitcoin Vs Fiat, How to trade safely, Is money root of all evil? Why did you have faith in Bitcoin?
After that, a lot of spammy users come and post tons of shit posts because these subjects are easy to abuse. So, the scheme is: One creates spammy thread, hundreds of spammy users come and spam. It's like, one user feeds other users to earn money.

This user called Susano is a perfect example of what I talk about right now.

So, I think, our priority should be to eliminate these type of users who feed spammers. Because of these users, we also catch spammers in their threads. Usually, these users and their spammy posters are the ones who use an AI.

And to my mind, if you look at user's post history, it's easy to understand whether they use an AI or not. When they use an AI, their posts and answers are general, you can feel that the answer is only answered according to thread title and it had no clue about the OPs post.

ChatGPT can only analyze english generated text at the moment, and it can't detect auto-generated text translated from one language to another (and possibly back again) using something like Google Translate.


Well, look at our website layout. Considering that it looks like a 2005 relic, I don't see how any advancements are going to change life around here (barring absolutely necessary additions like CAPTCHAs and CloudFlare).

To test level  of scrutiny of AI-text-detection-tool developed by openai I have fed it with ChatGPT-generated- text which was acknowledged by one of forum member.



and got the following responce:




To whom should I trust?

For me, I think AI generated text would sound almost too perfect with tenses and wordings to begin with. It would also touch every point as in a summary of what text books chapters or web pages would provide.
No one is beyond mistakes, but most times these human text points can even be made better with the use of text apps like Grammarly to correct errors and appropriate tenses in order. Hence, why it is reviewed on second note.
Unless one is foolish enough to not write on a few points, wherein more details is inputted than necessary, it could be easily detected as AI generated. Thank God for AI text detection tools, else who could really note the difference.

The original old, dusty, hole-punched books can be consulted thousands of years from now thanks to digitization. To preserve something you don't have to keep it in its original format.

Besides I think they are two different things. I would leave the forum format as it is. However, for innovations such as IA, I am clear, if it were up to me all text that is a mixture of IA with human content and results in a quality post, welcome.

Yes but TBH I kind of like it that way. I don't think the forum should be overhauled -- especially the design. The forum's primary purpose now is to act as a repository / archive for satoshi's posts, along with Hal and a handful of other early contributors / devs. As well as anything that could be regarded as "historically significant" in the timeline of bitcoin or cryptocurrency in general. It really doesn't matter if thick layers of spam get added on top so long as those early, formative posts are still visible an intact. And it would be weird to view satoshi posts in a new forum design that wasn't around when he was here.

The forum is staying in the time of the Australopithecus, going hunting with a club, while the world is on its way to the fifth industrial revolution.


AI is being used to improve the work in various fields right now, and the change it is generating is among the fastest in history. Those who worry that AI will jeopardize their income from signature campaign would do better to worry that AI will make this forum obsolete if they insist on fighting it instead of embracing it.

OK well I'm prepared to concede that it matters more whether a post is spam rather than if it was written by AI.

Just seems like cheating, and for the most part they are low-information posts that don't really add anything substantial to the conversation. At least they are well-written, lol.

Most people using ChatGPT to post on the forum wouldn't be writing anything of quality regardless, pumping out posts that are usually just coherent & on-topic enough to not be deleted as spam.

So basically we're back to where we were before.

Spammers gonna spam, AI or no.

Forum is  on the edge of collision with outside world.

While Elsevier  has published scientific paper writen in  coauthor with ChatGPT,    forum tends to ban any outcome from this AI.

A possible "out" my be the relevant  reference to source of content published on forum if this content came from  ChatGPT, IMHO

I am on verge of becoming anti-AI.
In hands of malicious people this can be used as a weapon.

AI crap generation is getting seriously dangerous, now ChatGPT created AI-generated version of Joe Rogan podcast, you can listen entire podcast on youtube, it's around 50 minutes long.
Joe Rogan is obviously concerned where all this AI experiments are going and he tweeted that this is going to get very slippery.
If they can generate this realistic audio podcast AI conversation than they sure can generate AI forum conversation, and I wonder what is next, Skynet, Terminators, etc.  
I really don't know how to identify that something is AI after having reasonable conversation with it, but maybe asking controversial questions can help to identify AI.
Joe Rogan AI Experience:
https://www.youtube.com/watch?v=meu0CoYv3z8

I disagree. A lot of people nowadays are using AI-powered writing assistants like spell and grammar correction tools or AI-powered translation tools to enhance their writing skills. Therefore, it has become quite challenging to tell apart content that's been generated by AI and content that's been aided by AI without taking into account the wider context.


I don't completely agree with that either. They surely can give acceptable and relevant responses to specific questions. The areas where AI language models may fall behind is understanding the bigger picture of the conversation or interpreting the emotional tone of the text. This is where human intuition and comprehension still play a crucial role.

I think you are confusing between AI generative language models and "simpler" pre-trained chat bots or voice assistants. While both use AI technology, generative language models like ChatGPT are more advanced and capable of producing complex and creative language outputs, whereas pre-trained chatbots typically rely on scripted responses to specific queries or commands.