Good Initiative by Royse777!

[joker_josue]

Has anyone ever thought of asking ChatGPT to make a script with the aim of detecting whether the text is written by the chat or not? 

It would be fun to know the answer, and to know if the supposed script would work.

[1714773534]

1714773534

[1714773534]

1714773534

[1714773534]

1714773534

[1714773534]

1714773534

[Hispo]

Good initiative but I am afraid it could be just matter of time before the first wrongly accused campaign participants appear and in the end the campaign manager will have the last word on whether a person stays or leaves with the presented evidence.

I'd wish that detecting AI generated content was as easy as a Bitcoin address signature.
This kind of technology is seemingly going so fast that security measures are staying far behind. 

[libert19]

Sounds spot on, it'd thing to see how people will recognize such texts. Hoping ai users will be revealed in public so other people can also learn how to recognize such texts and also malicious actors learn how to stay undetected 

[Lucius]

~snip~

I think that you are very confused and that you are looking for something that is impossible to achieve at this moment. You can play with these detectors as much as you want, but I don't consider them reliable enough to kick someone out of the campaign based on such detections.

The initiative is good, its implementation is very questionable, and I'm not even talking about the time that the manager will have to spend to assess whether something is original content or AI generated. However, if you think you can help the forum in detecting those who abuse it, you can always click on the "report to moderator" button and give your contribution

[suchmoon]

Has anyone ever thought of asking ChatGPT to make a script with the aim of detecting whether the text is written by the chat or not? 

It would be fun to know the answer, and to know if the supposed script would work.

https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text/

However:

Our classifier is not fully reliable. In our evaluations on a “challenge set” of English texts, our classifier correctly identifies 26% of AI-written text (true positives) as “likely AI-written,” while incorrectly labeling human-written text as AI-written 9% of the time (false positives). Our classifier’s reliability typically improves as the length of the input text increases.

Having said that...

You can play with these detectors as much as you want, but I don't consider them reliable enough to kick someone out of the campaign based on such detections.

If someone makes 20 posts and most of them are labelled as "likely AI" by multiple detection tools and generally sound like vague bullshit, then... you know... it's a duck.

Just like with any other issue, e.g. plagiarism - you don't rely on a tool 100%, you need to look into the context of how this occurred, what the potential intent was, etc.

[Lucius]

If someone makes 20 posts and most of them are labelled as "likely AI" by multiple detection tools and generally sound like vague bullshit, then... you know... it's a duck.

I can agree on that, but considering that the majority will say that any free work is a waste of time, I wonder who will spend their time hunting this new generation of AI posters if we also take into account the attitude of the moderators towards such reports. It seems to me that the majority of such reports could remain unhandled.

Just like with any other issue, e.g. plagiarism - you don't rely on a tool 100%, you need to look into the context of how this occurred, what the potential intent was, etc.

Actually, I always rely on some tool that can detect plagiarism, and that tool gives me possible sources that I then check and try to determine if someone has plagiarized their post, whether it is perhaps that someone is the original author and forgot to post the source. I always approach such matters very seriously, and even if I make a mistake, I know that the moderators will not make their decision hastily.

[JoyMarsha]

I think Royse's effort to limit the usage of ChatGPT by forum participants in a signature campaign she manages(Yolmix.io) is a smart move. Since there is no developed or well-known tool for such detection, the issue will now be how to be able to detect anyone using the ChatGPT to make posts. The more confusing aspect of utilizing ChatGPT is that it responds to questions presented to her in a completely different way(What a user asks to get his answer will be different from what other user asked to get her answers)

I'm really interested to see how the forum members report on ChatGPT turns out.

[AB de Royse777]

I am sorry, everyone had a wrong assumption. I don't deserve the credit but it's Stake manager Carollzinha. I was inspired from his/her post.

Actually not only some of our campaign members were using AI to generate their posts (and they had a payment cut for such behavior) but a lot of other forum members have been found using it as well. As Stake Campaign Manager all I can do is to include it as a rule in our campaign and hope the other Campaign Managers will do the same, for the forum health

As LoyceV said, the idea of rewarding was already existed too. What I did is - learnt from the past experience and mixed it with the new idea to improve the service we provide. It will be challenging but it's needed to appreciate the posters who do good job and genuine.

[Mpamaegbu]

If you want a good laugh, see here.

I had a good laugh at that. It's crazy. What was the poster actually thinking a wallet was. Well, I better stick to their advice and get a conditioner to be cleaning my wallet 🤦

And I expect that more and more of the long 0-value posts be reported as AI texts if they're written proper English (which may or may not be correct).

I think a more drastic step should be taken and not only reporting such Al posts. Those accounts should be tagged. It's deceit for me. Except if such account is registered in a way that the community recognizes it's an Al account. We've to know when we're exchanging correspondences with scripts so we don't throw in all our emotions into such banters. Computers don't have emotions.

[Cryptomiles1]

Well to me this is a good initiatives by Royse777, If i must go deeper it can only be common with bounty campaign instead of signature campaign since there are no much participants. But in normal bounty campaign there are over Hundreds or a thousand (if there's no limitations) participants which might join the campaign and if any of them are bot or using IA to write a report it would be very hard to detect.
In essence what are my saying, this principles should also be implemented in bounty's not only signature campaign and btw, there are limited slot for sig campaign rather than bounties. Most times whenever i made out chance to visit some bounty threads i kept wondering how some participants could just registered a bounty today and immediately submitted all reports without waiting for weekly deadline or even without trying to spread out their task within at least 4 to 6 working days of their signing up.

I don't know if this has already been discussed earlier.

[joker_josue]

As LoyceV said, the idea of rewarding was already existed too. What I did is - learnt from the past experience and mixed it with the new idea to improve the service we provide. It will be challenging but it's needed to appreciate the posters who do good job and genuine.

My suggestion is to allow users who suffer suspicions to justify themselves and show that the analyzes are wrong.

Otherwise, I think anyone who makes this type of post should be penalized.

[bullrun2024bro]

Although some posts may look like they are the work of AI at first glance, how will you prove that they are really the product of an AI Chat? I have tried 5-6 detectors and I am not satisfied with the results, because even very small changes in the entered text can give completely opposite results.

Here is the part of your post that one detector says is even 93% fake.

You're absolutely right and that's the tricky part.

Several European universities are conducting research on this topic and are currently testing a software for the automatic recognition of ai-generated essays or scientific articles. Unfortunately, as things stand, the results are anything but satisfactory. AI-generated texts like e.g. from ChatGPT can be modified by small adjustments in a way that a reliable error detection is not possible.

And it is the same with the various 3rd party AI trackers currently offered by many websites. These also do not work reliably and can often be deceived by adjusting just a few words so that a text that was previously output as completely AI-generated subsequently has an extremely low AI score.

Unfortunately, it will be quite difficult to make reasonable statements if someone really used AI to create forum posts. The easiest guess is probably with users who write qualitatively completely different posts from one day to the next - even if there is of course no 100% certainty here. Nevertheless, I think that just the fact that posts will be more closely checked in the future will lead to an increase in posting quality in general.

[NeuroticFish]

I think a more drastic step should be taken and not only reporting such Al posts. Those accounts should be tagged. It's deceit for me. Except if such account is registered in a way that the community recognizes it's an Al account. We've to know when we're exchanging correspondences with scripts so we don't throw in all our emotions into such banters. Computers don't have emotions.

The problem is that in many cases it's not easy to draw the line and tell beyond doubt that a post is AI generated or not.
And in such cases, while deleting one more more posts is not the end of the world, tagging an account will make it no longer be accepted into the reputable campaigns.

Plus, your proposal has a possible flaw*: if it's AI generated text, then the account can be directly reported for plagiarism and will be banned (thus making the tagging pointless).


* it's only "possible flaw" because it can take a lot until plagiarizing account actually gets banned, hence the tag can have a good effect, albeit for limited time.

[Volgastallion]

For me the real problem its with the exagerated IA post or accounts who are clearly one , and the idiots who still talk with that accounts.

Im refering that one who post something like you have to protect your wallet of leather or sintetic materials..... come on man. And i think if i not remember bad was Neuroticfish the man how catch that no sense and post that hilarious post.

[summonerrk]

I was going through the Services Child board and found that the Yo!Mix Bitcoin Mixer Signature Campaign was Open. As like everyone does I also went through it and found on the Campaign ANN a new initiative. This new initiative as per my knowledge has not be initiated by any other BM other than @Royse777. I think it is tricky but can get good results. Quoting what he wrote

To prevent AI domination effective from today we are introducing an incentive for forum users. Find AI written posts on this campaign
and report to me either in public or in forum PM, please be sure you have enough reference to support the claim. For successful report
the reporter will receive the weekly payment instead of the accused campaigner. The campaigner will be removed immediately.

I created this thread to know what you guys think about it. Do you feel he/she as in Royse777 will be able to catch those Bounty hunters who are using AI chatbots to earn Bitcoin on a signature campaign managed by him? Will other BMs follow his initiative in the future?

I think it is not impossible but can be made possible if someone knows how to find an AI script for a reply or a topic.

Royse777 absolutely right.
We all guess that in the near future, or already, the forum will be spammed with hundreds of messages authored by AI. And while the algorithms are not perfect, and often the AI writes in general incoherently, we can calculate unscrupulous authors.

Thats info what I have found:

How can a person distinguish an AI text
The neural network writes too concretely. For example, researchers from Google Brain came to the conclusion back in 2019 that AI uses the article the too often because it works on predicting the next word in a sentence. But it is clear that this feature is inherent only in English-language texts.
ChatGPT is already being used as a replacement for search engines, since it gives quite specific answers. But things are worse with her reasoning: for questions like "What is love?" the AI will give a dry formulation, and not an attempt to analyze a complex feeling.
Also, a neural network can very confidently write an absolute lie. It is not her nature to doubt.
The neural network omits the context and personalization of the text. ChatGPT will not be able to convey the experience of a particular person — only try to imitate the experience of millions of people on whose texts the model is trained. But such materials usually lack depth and author's feeling: it's one thing to write a template essay on a literary work, and another to tell how you spent the summer.
Also The neural network does not make mistakes. 

Also here is DETECTOR:
https://openai-openai-detector.hf.space/

[suzanne5223]

Has anyone ever thought of asking ChatGPT to make a script with the aim of detecting whether the text is written by the chat or not? 

It would be fun to know the answer, and to know if the supposed script would work.

I like the idea introduced by the user mentioned by the OP but ChatGPT and other Ai text or written composers are just a piece of plagiarism equipment. To my knowledge, there's no way you can detect whether the text is written by the chat because the artificial intelligence machine was programmed to only searches online all the available content base on the question provided. Besides, it works just like search engines.

[libert19]

I'd wish that detecting AI generated content was as easy as a Bitcoin address signature.

Tools to detect ai text are sure to come in matter of time. Where there is need, there is invention.

This kind of technology is seemingly going so fast that security measures are staying far behind.  

I have been seeing ai development since years, not sure if it's really that fast.

[joker_josue]

Tools to detect ai text are sure to come in matter of time. Where there is need, there is invention.

This will be very tricky. I won't say impossible, more difficult.

Because unless the AI leaves a pattern or signature in the text, it's going to be hard for a detector to tell whether or not it's an AI.
In addition, text AIs can even evolve to give errors in writing, to make it difficult to detect. Or be less "mayors" in the response.

[Little Mouse]

Tools to detect ai text are sure to come in matter of time. Where there is need, there is invention.

That's not going to be a free one of course. Everyone will come up with a paid subscription model. Allow users to generate text for free, and ask managers to check whether the texts are from AI or not for a fee. How innovative idea! lol. It's the same idea we used to see the corporate business most of the time.
However, it's not that easy to detect. I had a test with other tools and shared my views another day. It's not going to be that easy.

[Aanuoluwatofunmi]

I created this thread to know what you guys think about it. Do you feel he/she as in Royse777 will be able to catch those Bounty hunters who are using AI chatbots to earn Bitcoin on a signature campaign managed by him? Will other BMs follow his initiative in the future?

It's such a good move to work against the bounty cheaters on the forum, i think a manager has well announced on something related that when reported with valid evidence they will be rewarded a d the participant will be removed from their campaign and also tagged red, it's a choice if others finds a need for requesting on such or engage doing it by themselves, without stating that already we have some members always on a look for cheaters of any kinds on the forum and they always fish them out even though it may take time to discover any.

I think it is not impossible but can be made possible if someone knows how to find an AI script for a reply or a topic.

Yes, one needed to know how to find one.