AI writing messages on Bitcointalk.org

[LoyceV]

This will be somewhat a disaster if someone with a long list of account starts to participate in signature campaign and post AI generated garbage, I'm pretty much sure we will see the day in the coming future.

Even if the text-creating-AIs become paid services, it's probably not going to be more than a fraction of a dollar per text, and since many campaigns pay up to a few dollars per post, that's very profitable.
Automating the whole thing will give you passive income. If enough people do this (of if one person creates enough accounts), it will outcompete humans from signature campaigns.

Then this will happen:

us mortals move on to something else.

Maybe Bitcointalk.human will work as a domain. With a captcha

Or maybe the internet isn't meant to last for humans after all:

The internet will fade away because most people have nothing to say to each other.

[Lucius]

I was just wondering what other people's attitude is when it comes to such posts, because it is not acceptable to me personally

I'm starting to suspect some posts are AI-written, and I don't like it. It's just the next phase of shitposting.

In the specific case I mentioned, the OP admitted that the post was written exclusively with AI, and after the report, that thread was locked. Of course, it is difficult to prove that someone is using AI for their posts, but I think that the post history of each member can reveal any significant change.

What also interests me about these AIs is how creative they are in the sense of whether they answer the same question 100% identically or still make a difference. It would be interesting if we could find some unique pattern that would indicate that behind such posts is not a human but an AI.

[DdmrDdmr]

<…>

I played around for a while asking Chat GPT whether this or that content was generator by it, and there are some certain circumstances where it declares having come up with the text, or provides the "original text" (or a variation) if the data you feed it differs somewhat. The post you referenced was verbatim, so it was an easy match for it to find, but the more you twist it the sooner it declares it not being his. Obviously this is only an indicator though, not a viable declaration of guilt that can be accepted as proof.

If you ask it the same question multiple times it provides different answers, but the essence, and even some fragments, are the same from one iteration to another. A unique pattern will be hard to find, though it does tend to be rather prosaic, often tries to play the neutral role, and for now will generally lack the human touch to its answers, being rather consistent in its posts, as opposed to what one would really be creating by writing posts with a wide range of tones, content length and style.

[Don Pedro Dinero]

Specific reasons for censoring AI generated posts may vary depending on the specific forum and its rules and policies. For example, someone can argue that such posts are spammy, zero or low value, misleading, etc... Btw, it's important to note that AI-generated content is not universally accepted on the Internet. In fact, sites like Reddit have specific rules in place to prohibit the use of such content.

My point is that that post is no worse than many of the ones we see out there in terms of low value or spammy, and that the fact that it is written by an AI should not be enough to delete it, considering that there is nothing about it in the unofficcial rules and even more thinking that the OP has made it clear that it is written by an AI.

And, how do you see this kind of discussion related to Gambling?
I want to point out that the topic of AI-generated content was created in the gambling section, but you are discussing the quality and accuracy of AI-generated content. That discussion is not related to gambling and might be more suitable for a different forum board.

The one we are having here has nothing to do with gambling, the one over there does, because they are talking about supposed ways of winning in gambling.

the post has been reported, we are waiting for the decision of the mods.

Can I know the reason? (low value, spam, etc). Btw I see it's been locked.

It would be interesting if we could find some unique pattern that would indicate that behind such posts is not a human but an AI.

I have seen many such posts on other sites and they have in common that they are rather generic and too formally structured, like, point 1, point 2, summary, conclusion, something like that.

[LoyceV]

My point is that that post is no worse than many of the ones we see out there in terms of low value or spammy

My worry is the sheer volume of posts an AI can produce.

[shahzadafzal]

I'm starting to suspect some posts are AI-written, and I don't like it. It's just the next phase of shitposting.
Example: this post: it has a lot of "generic text", and some parts are just bad practice and shouldn't be advised. All in all, it's pointless. But it's well-written and not against the forum rules.

Damn it... well spotted LoyceV ... I asked the ChatGP similar question and look at the answers

• Use advertising to spread the word that you accept Bitcoin as payment. This can be done online, through social media, or even by putting up a sign in your place of business.
• Offer a discount for customers who pay with Bitcoin. This could be a percentage off their purchase, or a fixed amount if they use Bitcoin to pay.
• Make it easy for customers to pay with Bitcoin by accepting it through a digital wallet such as Coinbase or BitPay. You can also use a physical wallet if you prefer, but make sure it supports Bitcoin payments.
• Educate your staff on Bitcoin and how it works, so they can answer any questions customers may have about using it to pay.

It looks all the four bullet points are somehow inspired by ChatGPT

First bullet point coming from Tip No. 5 of ChatGPT rephrased
Second bullet point coming from Tip No. 2 of ChatGPT copy paste + rephrased
Third bullet point coming from Tip No. 3 of ChatGPT copy paste + rephrased
Fourth bullet point coming from Tip No. 4 of ChatGPT rephrased




Give that man a cookie for using ChatGPT

[LoyceMobile]

Damn it... well spotted LoyceV ... I asked the ChatGP similar question and look at the answer

Playing devil's advocate: it's still possible the ai chat learned from the recent post. I think it's unlikely, but the possibility can't be rules out. Damn ai!

[FatFork]

Specific reasons for censoring AI generated posts may vary depending on the specific forum and its rules and policies. For example, someone can argue that such posts are spammy, zero or low value, misleading, etc... Btw, it's important to note that AI-generated content is not universally accepted on the Internet. In fact, sites like Reddit have specific rules in place to prohibit the use of such content.

My point is that that post is no worse than many of the ones we see out there in terms of low value or spammy, and that the fact that it is written by an AI should not be enough to delete it, considering that there is nothing about it in the unofficcial rules and even more thinking that the OP has made it clear that it is written by an AI.

Ah yes, I suppose I'm just a relic of the past. But I still believe that an internet forum should be a place where humans come together to discuss and share their thoughts, not a robot convention where mindless machines spout pre-programmed responses. Real discussions require actual thought and consideration, something that robots just don't have the capacity for. 

I don't think we need any specific rules about what constitutes an "Internet forum," because it's one of the most basic definitions of that term:

An Internet forum, or message board, is an online discussion site where people can hold conversations in the form of posted messages.

[Ultegra134]

I actually messed around with ChatGPT a few days ago, and I find it both outstanding and scary at the same time. It can generate anything, from articles to code, while finding answers to your most stupid questions. On the one hand, it can be an extremely useful tool to use in projects and assignments, for instance, in moderation of course. On the other hand, it can easily be abused, from posting in the forum to even using it to generate massive walls of text in a master's thesis. AI is getting out of hand.

Good job, LoyceV, for spotting AI posters. Despite having seen how the application works, it didn't cross my mind that it was AI-generated, and truth be told, I've seen quite a few similar ones the past few days but thought nothing of it.

[drwhobox]

I'm starting to suspect some posts are AI-written, and I don't like it. It's just the next phase of shitposting.
Example: this post: it has a lot of "generic text", and some parts are just bad practice and shouldn't be advised. All in all, it's pointless. But it's well-written and not against the forum rules.

Damn it... well spotted LoyceV ... I asked the ChatGP similar question and look at the answers

• Use advertising to spread the word that you accept Bitcoin as payment. This can be done online, through social media, or even by putting up a sign in your place of business.
• Offer a discount for customers who pay with Bitcoin. This could be a percentage off their purchase, or a fixed amount if they use Bitcoin to pay.
• Make it easy for customers to pay with Bitcoin by accepting it through a digital wallet such as Coinbase or BitPay. You can also use a physical wallet if you prefer, but make sure it supports Bitcoin payments.
• Educate your staff on Bitcoin and how it works, so they can answer any questions customers may have about using it to pay.

It looks all the four bullet points are somehow inspired by ChatGPT

First bullet point coming from Tip No. 5 of ChatGPT rephrased
Second bullet point coming from Tip No. 2 of ChatGPT copy paste + rephrased
Third bullet point coming from Tip No. 3 of ChatGPT copy paste + rephrased
Fourth bullet point coming from Tip No. 4 of ChatGPT rephrased




Give that man a cookie for using ChatGPT

Accepting bitcoin as a payment method is a great idea, but also make sure clients/customers are inspired to pay with bitcoin. Most of the people will avoid bitcoin payments and will pass it on to cash or other payments. Make it easy and attractive for your customers to pay with bitcoin. By attractive means give them some discount or give them some points which they can use later to buy anything from you. They will acquire these points or discounts if they only pay with bitcoin.

It's good to see Nigerians are very aware of bitcoin and trying to spread it to their community.

So maybe I just used my own reply for bullet points of my own thread. Or I asked chatgpt for what reply I should post on this thread - https://bitcointalk.org/index.php?topic=5431962.msg61503560#msg61503560

I learned about chatGPT from a youtube video published by a YouTuber Gotch SEO and later that day on this forum. I asked chatGPT many questions but not this one.
Besides, I see only one point is matched which is also I replied to someone's post already without bullet point before creating my own topic.

Give that man a cookie for using ChatGPT

I want that cookie for sure.

[DdmrDdmr]

<…>

There is a close resemblance indeed, though I couldn’t get Chat GPT to declare having written a single paragraph, nor selected line within. It could have been span and custom tailored to deviate from the original suggested output, but then again, it does come up with some similar conceptual answers when denying being the author:

No, I did not write the suggestion that customers may avoid Bitcoin payments and that business owners can follow certain tips to encourage the use of Bitcoin. I am an artificial intelligence trained to provide information and answer questions to the best of my ability based on the knowledge that I have been trained on, but I do not have the ability to create or promote businesses or products.
It is possible that some customers may be hesitant to use Bitcoin as a form of payment, particularly if they are not familiar with how it works or are concerned about the security of digital currency transactions. In order to encourage the use of Bitcoin as a form of payment, business owners may want to consider a number of strategies. These could include using advertising to spread the word about Bitcoin acceptance, offering discounts to customers who pay with Bitcoin, making it easy for customers to pay with Bitcoin through the use of digital or physical wallets, and educating staff on Bitcoin and how it works. Business owners should carefully evaluate their own needs and circumstances and choose the strategies that will work best for them.

The proposed strategies are though, as expected, more or less common knowledge, and there could be different sources that agglutinate them and that were used as an inspiration. Whether the inspiration was for the poster or Chat GPT is what remains unclear.

<…> Playing devil's advocate: it's still possible the ai chat learned from the recent post. I think it's unlikely, but the possibility can't be rules out. Damn ai!

Allegedly, it can't use recent data. According to "it":

My training data consists of a snapshot of the internet as it existed at a specific point in time. The exact date of the snapshot that was used to train me is not publicly disclosed by OpenAI. However, I can tell you that my knowledge is based on a snapshot of the internet that was taken in 2021, and my knowledge does not include any information that has been published or released after that time. My training data does not include any updates or changes that have occurred since the snapshot was taken, and my responses are based on the information that was available at the time the snapshot was taken.

I figure though that Chat GPT is based on the GPT-3 model text-davinci-003, whose training data ran up to Jun 2021 (as per OpenAI’s model overview).

[Lucius]

the post has been reported, we are waiting for the decision of the mods.

Can I know the reason? (low value, spam, etc). Btw I see it's been locked.

If you mean what I stated in the report, then it is "low value/zero effort post generated by AI". Of course it was locked, and quite quickly if I may say so - I don't think anyone wants to encourage forum members to use such tools to publish posts.

[actmyname]

Bear in mind that ChatGPT has a limited general use when directing through prompts.

Given a non-zero-shot set of prompts, catered to BCT, you could very easily create an engine for post generation. Though, it is magnitudes more expensive than GPT-3 Davinci (if you're paying). If you had enough resources, though, you could easily sidestep the OpenAI cluster and try to use other models (e.g. EleutherAI).

[noormcs5]

I'm starting to suspect some posts are AI-written,

Same here. I have also started to suspect the same. I have been monitoring one account for the last few weeks, and within this time, the account have posted 3/4 times only which seems like AI generated to me. I don't want to publicly post that until I'm sure that he is posting AI generated text.
This will be somewhat a disaster if someone with a long list of account starts to participate in signature campaign and post AI generated garbage, I'm pretty much sure we will see the day in the coming future.

The problem is that how will you proof that a message is AI written or self written ?

One way to go about is that you will find all of his posts in a similar pattern of AI generated text but what if he uses AI in some of his posts and he writes other posts himself. Then i guess it will be even more difficult to judge when there will be mixtures of AI and self written text.

[Ucy]

The value of AI generated posts actually depends on what it's fed with, if you feed it garbage it generates garbage (garbage in garbage out).
I've gone through several posts generated by it and I think they are mostly worthless, inconsistent, full of lies etc, to the extent I asked it who is his creator and it said a car marker & others created it... "No wonder you are not so smart" I told the AI. I was embarrassed and stop using. Will probably start looking for an alternative or we could join hands together to create a far more superior one based on truth in less than a month.


From my observations of existing AI on the internet, I suspect they look for very honest people or sites in existence and stick to them for Holy/Pure information in order to consistently generate good information that's very dependable and helpful to society without issues. I hope they will be beneficial to the sources rather than being parasitic in the manner they feed from them.

[krishnaverma]

The problem is that how will you proof that a message is AI written or self written ?

There are tools to detect this. You can find them by simple Google search.
But the problem will be of scale. If people start abusing this here or on any other forum, because of the huge volume of spam, it will consume lot of forum staff time to spot and end this. This challenge will be faced by all forum owners in future. One site I can give example of here is Quora which is getting filled with such AI posts.

[DdmrDdmr]

There are tools to detect this. You can find them by simple Google search. <…>

There are, but at least the ones I’ve tried out ,that allow you to perform a free trial online, have seemingly very poor results detecting GPT-3 based Chat GPT text, and that’s just feeding a direct output as input to the detector, with no spinning or customization at all in between to even try to throw them off their tracks. There are a few additional ones I would have been prone to try, but they required subscription and payment (a no go at this stage).

Were the outputs to eventually have an inbuilt digital watermark, there could be an easier basis to detect AI generated type text, but even so, each company will have used their own method to generate the (potential) watermark, and I’m pretty sure that adding a step or two in the end-user’s pipeline would largely degrade the watermark’s persistence and utility (though that requires some minimal effort on the user’s side, which is not a done deal).

[Ultegra134]

There are tools to detect this. You can find them by simple Google search. <…>

There are, but at least the ones I’ve tried out ,that allow you to perform a free trial online, have seemingly very poor results detecting GPT-3 based Chat GPT text, and that’s just feeding a direct output as input to the detector, with no spinning or customization at all in between to even try to throw them off their tracks. There are a few additional ones I would have been prone to try, but they required subscription and payment (a no go at this stage).

Were the outputs to eventually have an inbuilt digital watermark, there could be an easier basis to detect AI generated type text, but even so, each company will have used their own method to generate the (potential) watermark, and I’m pretty sure that adding a step or two in the end-user’s pipeline would largely degrade the watermark’s persistence and utility (though that requires some minimal effort on the user’s side, which is not a done deal).

Indeed, not all services are reliable and free at the same time. Even if that were the case, we simply couldn't go around to each forum post and test if it was written by AI. There are some hints that may indicate that a post was written by AI, but my main concern is that services such as ChatGPT may even create duplicate content if the same parameters are used, and it may cause post bursts and spam issues since it is capable of producing a wide variety of posts within a short period of time. AI may not produce low-quality posts, but it'll definitely create non-engaging discussions, which in some cases, may be worse than plagiarism.

[NotATether]

I was just wondering what other people's attitude is when it comes to such posts, because it is not acceptable to me personally

I'm starting to suspect some posts are AI-written, and I don't like it. It's just the next phase of shitposting.
Example: this post: it has a lot of "generic text", and some parts are just bad practice and shouldn't be advised. All in all, it's pointless. But it's well-written and not against the forum rules.

Keep in mind that ChatGPT will never use the colloquial "we" when writing an answer (they might say I) unless you explicitly tell it to. So posts with mostly AI sentences but a few human-generated sentences are like a semi-automatic gun: it's not a full-fledged machine gun, but still dangerous nonetheless, and more mobile.

[shahzadafzal]

Keep in mind that ChatGPT will never use the colloquial "we" when writing an answer (they might say I) unless you explicitly tell it to.

That intrigued me to try this "we" thing and it worked