Auto detect Scams, Spam and AI in the forum

[seoincorporation]

The forum could use AI to fight against AI, Scams and Spam.

With the right tools it could be done, but isn't an easy task, The fact that i was able to spam with AI for multiple days is a good example bout the current problem that we have on the detection systems. It was a user who reported the post while i was expecting to get busted by the staff.

So, the thing is:
1.- we as users can create tools for monitoring the forum post and find who is exhibiting bad behavior and report it to mods or in the respective threads.
2.- Mods could have better tools for their mods job.
3.- The forum could implement some integrations to the code with tools like GptZero and any IA model for the scams and spam detection.

What i have right now.

With a python script i can choose a link and get all the posts in a JSON:

Code:

from bs4 import BeautifulSoup
import json
import sys
import requests
import re

def extract_after(text, key):
    try:
        return text.split(key)[1].split()[0]
    except:
        return None

def parse_quote_header(header_text):
    match = re.search(r"Quote from:\s*(.+?)\s+on\s+(.*)", header_text)
    if match:
        return match.group(1).strip(), match.group(2).strip()
    return None, None

def extract_user_profiles(soup):
    profiles = {}
    for td in soup.find_all("td", class_="poster_info"):
        a = td.find("a")
        if a:
            name = a.text.strip()
            href = a.get("href")
            profiles[name] = href
    return profiles

def extract_quotes_recursive(container, user_profiles):
    quotes = []
    headers = container.find_all("div", class_="quoteheader", recursive=False)

    for header in headers:
        quote = {}
        link_tag = header.find("a")
        quote["link"] = link_tag["href"] if link_tag else None
        user, date = parse_quote_header(header.get_text(strip=True))

        quote["author"] = user
        quote["profile_url"] = user_profiles.get(user, None)
        quote["date"] = date

        quote_block = header.find_next_sibling("div", class_="quote")
        if quote_block:
            quote["quotes"] = extract_quotes_recursive(quote_block, user_profiles)
            for q in quote_block.find_all("div", class_="quote", recursive=False):
                q.decompose()
            quote["content"] = quote_block.get_text(strip=True)
            quote_block.decompose()
        else:
            quote["quotes"] = []
            quote["content"] = ""

        header.decompose()
        quotes.append(quote)

    return quotes

def parse_html_posts(html_content):
    soup = BeautifulSoup(html_content, "html.parser")
    post_blocks = soup.find_all("td", class_="msgcl1")
    user_profiles = extract_user_profiles(soup)
    posts_data = []

    for block in post_blocks:
        post = {}
        anchor = block.find("a")
        post["message_id"] = anchor.get("name") if anchor else None

        poster_td = block.find("td", class_="poster_info")
        if poster_td:
            user_link = poster_td.find("a")
            post["author"] = user_link.text.strip() if user_link else None
            post["profile_url"] = user_link["href"] if user_link else None

            activity_text = poster_td.get_text()
            post["activity"] = extract_after(activity_text, "Activity:")
            post["merit"] = extract_after(activity_text, "Merit:")

        subject_div = block.find("div", class_="subject")
        post["title"] = subject_div.get_text(strip=True) if subject_div else None

        date_div = subject_div.find_next_sibling("div") if subject_div else None
        post["date"] = date_div.get_text(strip=True) if date_div else None

        post_div = block.find("div", class_="post")
        if post_div:
            post["quotes"] = extract_quotes_recursive(post_div, user_profiles)
            post["content"] = post_div.get_text(strip=True)

        posts_data.append(post)

    return posts_data

def main():
    if len(sys.argv) < 2:
        print("Usage: python3 post_last.py <URL> [output.json]")
        sys.exit(1)

    url = sys.argv[1]
    output_path = sys.argv[2] if len(sys.argv) > 2 else "bitcointalk_parsed.json"

    try:
        headers = {"User-Agent": "Mozilla/5.0"}
        response = requests.get(url, headers=headers)
        response.raise_for_status()

        posts_json = parse_html_posts(response.text)
        with open(output_path, "w", encoding="utf-8") as outfile:
            json.dump(posts_json, outfile, indent=2, ensure_ascii=False)

        print(f"Success! Saved to {output_path}")

    except requests.RequestException as e:
        print(f"Error fetching URL: {e}")
        sys.exit(1)

if __name__ == "__main__":
    main()

Code:

$ python3 posts.py https://bitcointalk.org/index.php?topic=5547609.msg65510556#msg65510556 out.json

If we send that JSON yo an AI Agent to analyze the data, we could get a report from each post, i trained my agent to post the feedback in the next format:

And that shows us a basic example about how we could use AI to create a nice filter and avoid the bad guys.

Some implementations that i would do to my code would be:

-Get only the last post and not all the thread
--This could be done by having the sections' links and making some sorting based on the date and time, for this we don't need AI
-Add IA detectors API and not only ChatGPT

The way that i think this should work:

This would work better if it were implemented on the server side and not on the user's side, it would be easier to verify the post when the users push the post button than waiting for the automation to find the new post.

And my question for the community is: Is it really worth trying to automate a process like this, or are we fine with the current situation?

[LoyceV]

The fact that i was able to spam with AI for multiple days is a good example bout the current problem that we have on the detection systems.

The real problem is not instantly banning spammers like yourself. It doesn't matter if it takes a few days, as long as the punishment is severe enough.

[Free Market Capitalist]

The real problem is not instantly banning spammers like yourself. It doesn't matter if it takes a few days, as long as the punishment is severe enough.

I don't see you very thick skinned with this topic.

The forum could use AI to fight against AI, Scams and Spam.

I imagine it's like the anti-plagiarism tools you are proposing. Internet made plagiarism easier and at the same time plagiarism detectors emerged.

And my question for the community is: Is it really worth trying to automate a process like this, or are we fine with the current situation?

I think you will be more successful if you present something that works and make it available to the community. After you have been posting in a paid campaign with your AI agent without disclosing anything until you get caught if you want to present a solution you better present a solution that is real and works. You are going to have more chances of good reception by the community. You've already seen the first response.

Then, assuming it works, we'll have to see if it gives false positives and such.

[LoyceV]

I don't see you very thick skinned with this topic.

What does that even mean? A thick skin against spam? I've been following AI Spam Report Reference Thread, and I think the forum is far too lenient on chatbot spammers. It's easy to spam, hard to detect, and without strong punishment, the spammers can actually win this.

The forum could use AI to fight against AI, Scams and Spam.

I imagine it's like the anti-plagiarism tools you are proposing. Internet made plagiarism easier and at the same time plagiarism detectors emerged.

I wouldn't be surprised if the AI-companies end up earning from both sides: earn from creating spam, and earn again from detecting their own spam. It's like creating virusses to sell your virus scanner.

[Churchillvv]

I wouldn't be surprised if the AI-companies end up earning from both sides: earn from creating spam, and earn again from detecting their own spam. It's like creating virusses to sell your virus scanner.

I think it's called monopoly or Cannibalization and that's exactly the point why this AI companies are busy creating new tools to still maintain it's position in the market. You're right! they are profiting from the both side.
To OP don't you think it makes zero sense to fight something with the same thing? you want to fight AI spam with AI detector.

The solution you offer is already stereotype or judged before you even posted it, as long as you already have an issue whatever you offer now as solution will be sense as a means to justify your actions.

[GeorgeJohn]

And my question for the community is: Is it really worth trying to automate a process like this, or are we fine with the current situation?

This solution is what you would have done when your weren't using artificial intelligent [AI] to make a post, now you are a victim of Al user and nobody would like to take this your idea serious

[ABCbits]

The fact that i was able to spam with AI for multiple days is a good example bout the current problem that we have on the detection systems. It was a user who reported the post while i was expecting to get busted by the staff.

Staff doesn't have that much resource to monitor all created posts/threads. And AFAIK, the only automated detection and removal system is MindlessElectron bot[1].

And my question for the community is: Is it really worth trying to automate a process like this, or are we fine with the current situation?

It comes down to whether how good is the detection, including how frequently false positive is happening. And as reminder, this forum doesn't moderate scam. Besides, admin could do something much more simple by updating rule to make spammer got permanently banned/nuked much faster, especially one who use AI/chatbot to create spam.

[1] https://bitcointalk.org/index.php?action=profile;u=1136003

[Free Market Capitalist]

I don't see you very thick skinned with this topic.

What does that even mean?

It means that I see a viscerality in this issue that stands out over your usual rationality.

A thick skin against spam? I've been following AI Spam Report Reference Thread, and I think the forum is far too lenient on chatbot spammers.

Perhaps the forum does not share that visceral vision.

And don't you think the thread itself is pretty lenient? According to the rules of the thread an account could create generated AI posts, tweak them a bit himself and as long as they were detected below 70% generated AI he shouldn't be tagged. I guess spammers don't do it because they are too lazy.

It's easy to spam, hard to detect, and without strong punishment, the spammers can actually win this.

Win what? Spammers are too dumb to be able to generate premium content using AI.

[LoyceV]

It means that I see a viscerality in this issue that stands out over your usual rationality.

Maybe. I'm still feeling ill, makes me wonder if that does in fact change my view on things.

And don't you think the thread itself is pretty lenient? According to the rules of the thread an account could create generated AI posts, tweak them a bit himself and as long as they were detected below 70% generated AI he shouldn't be tagged.

I don't think that's because of being lenient, it's because detection tools aren't perfect.

[hugeblack]

A while ago, I remember an account using a plagiarism checker tool and report posts, hundreds of accounts were banned because of it. So, if the forum wants to be strict about spam, such tools would be useful.

In the past, Mining Board didn't allow spam, so a minimal amount of spam might be a good thing (let's not forget that interest in the forum is starting to wane).

[philipma1957]

A while ago, I remember an account using a plagiarism checker tool and report posts, hundreds of accounts were banned because of it. So, if the forum wants to be strict about spam, such tools would be useful.

In the past, Mining Board didn't allow spam, so a minimal amount of spam might be a good thing (let's not forget that interest in the forum is starting to wane).

Asic mining has become a huge business or tiny little gear and maybe a few space heaters.

gpu mining is essentially dead.
☠️

sooo forum lost two large content sections.

[Lucius]

The forum could use AI to fight against AI, Scams and Spam.

With the right tools it could be done, but isn't an easy task, The fact that i was able to spam with AI for multiple days is a good example bout the current problem that we have on the detection systems. It was a user who reported the post while i was expecting to get busted by the staff.

I don't know if the mods and staff have ever hunted down offenders on the forum, but what I do know is that they act on user reports, so I don't know why you're surprised that the staff didn't catch you, because that's not their job on the forum (in the literal sense of the word).

Besides what you did "for days", some others do it for weeks or even months even if someone reports them - except for newbies who are easy to nuke, the rest have a lot of credits before they get a permanent ban.

And my question for the community is: Is it really worth trying to automate a process like this, or are we fine with the current situation?

Even if you automate the process, you must have people who will additionally check any possible abuse of AI (perhaps the same ones who caught you), and then finally the mods to whom such posts must be reported and who will decide on the punishment. Sometimes the same process will need to be repeated for the same user 10 or 30 times before they are permanently banned, and now you can probably understand how much time would be needed to invest in such an operation.

[uchegod-21]

In your poll, I could not vote because the options are limited. If the process is automated, who will use the tool; any forum member or only the moderators. I'll suggest only moderators have it to avoid unnecessary dramas. Such that when a particular user is reported to moderators for spamming, Ai spamming or plagerism, they will have to check the profile for previous actions and draw a nice decision.
Op, can you run the profile of LoyceV with your bot? I want to see his score, I heard he is half human, half bolt (winks)

[Alone055]

I wouldn't be surprised if the AI-companies end up earning from both sides: earn from creating spam, and earn again from detecting their own spam. It's like creating virusses to sell your virus scanner.

It's already happening:



Quillbot is a website that allows you to generate AI texts, has an AI text detector, and also allows you to humanize AI-generated content, lol.

[yahoo62278]

I wouldn't be surprised if the AI-companies end up earning from both sides: earn from creating spam, and earn again from detecting their own spam. It's like creating virusses to sell your virus scanner.

It's already happening:



Quillbot is a website that allows you to generate AI texts, has an AI text detector, and also allows you to humanize AI-generated content, lol.

I don't feel like people are looking at the long game here as far as AI and this forum is concerned. Yes you can generate posts, yes you could build some sort of detector, yes it probably has some good things. The issue is what happens when this forum is just AI generated posts? All the real users dropping knowledge and helping bitcoin grow disappear. All the companies advertising see it as a spam haven with no real users. Forum starts slowly dying and all these people that were too lazy to make an actual contribution and only thought about making a few bucks will have to look for a new income source and 1 that likely pays a lot less.

So While I think detecting AI and getting rid of those that are using it here on the forum would be a great idea, I also think it won't happen and my above concern will happen.

[nutildah]

humanize AI-generated content

This is one of the biggest misnomers to be applied to any type of new technology. The idea that a machine attempting to emulate a human being can further "humanize" its output is absurd. The machine is doing all the work -- there is no human beings involved in "humanization." Thus a more proper term would be "shitifier" as it makes output more shitty albeit in a very robotic way.

As far as the subject of this thread is concerned, I don't think the mods want more work of having to go out of their way to determine what is or isn't AI content. It is already hard enough to get some posters to admit they are using AI. The ultimate offense that determines if a post should be removed is whether or not it falls into the category of "spam" in the eyes of the moderators.

[philipma1957]

unfortunately i can not answer the poll since my honest answer is I don’t know what is better.

you left out that choice.

as a suggestion do it for a month and see the results.

[LoyceV]

This is one of the biggest misnomers to be applied to any type of new technology. The idea that a machine attempting to emulate a human being can further "humanize" its output is absurd. The machine is doing all the work -- there is no human beings involved in "humanization." Thus a more proper term would be "shitifier" as it makes output more shitty albeit in a very robotic way.

I can't find back the post, but I saw someone describe it as posts having a metalic taste. You don't need AI detectors to know something's off with a post, but it makes it easier to be convincing.
Most of the time I can't get myself to read AI output. It just feels wrong.

[Alone055]

I don't feel like people are looking at the long game here as far as AI and this forum is concerned. Yes you can generate posts, yes you could build some sort of detector, yes it probably has some good things. The issue is what happens when this forum is just AI generated posts? All the real users dropping knowledge and helping bitcoin grow disappear. All the companies advertising see it as a spam haven with no real users. Forum starts slowly dying and all these people that were too lazy to make an actual contribution and only thought about making a few bucks will have to look for a new income source and 1 that likely pays a lot less.

Those who are doing all that don't give a f**k about this forum and the value it holds. A person generating his posts through an AI only to save a few minutes while earning money from the paid signature he's wearing barely cares that he's contributing towards ruining a place where Bitcoin was born or its creator(s) were once present, and that how valuable and memorable this place is. They just don't care.

This is one of the biggest misnomers to be applied to any type of new technology. The idea that a machine attempting to emulate a human being can further "humanize" its output is absurd. The machine is doing all the work -- there is no human beings involved in "humanization." Thus a more proper term would be "shitifier" as it makes output more shitty albeit in a very robotic way.

You're right, but as a wild guess, I believe so many people use these tools even in this forum to bypass detectors and get marked as AI spammers.

Since a lot of experiments are happening these days, why not try it and see how it does?  

I'll generate a post using ChatGPT, run it through AI detectors, then humanize the text using the humanizer from Quillbot, and then run it through the detectors again, and then I'll add the outcome of both tests below. I'll be using your post (the one I'm responding to) to generate a response for this test.

Response generated through ChatGPT:

Thank you for sharing your viewpoint. While it is understandable to question the use of the term "humanize" in relation to machine-generated content, it is commonly used to describe efforts made to adjust AI output so it resembles natural human language more closely. This process does not involve actual human emotion or intent, but rather algorithmic adjustments to tone, structure, and phrasing.

Regarding forum moderation, your observation is valid. Identifying AI-generated content can be challenging, especially when users do not disclose their use of such tools. Therefore, it is logical that moderators prioritize content quality and relevance over the method by which the post was created. Posts deemed as spam, regardless of origin, are more likely to be subject to removal.

GPTZero: 100% AI-generated
Sapling.ai: 100% Fake
Copyleaks: 100% AI Content Found
Quillbot: 100% AI-generated

Humanized version using Quillbot's AI Humanizer:

We appreciate you offering your point of view. Although it is reasonable to doubt the use of the term "humanize" in reference to machine-generated content, it is often used to characterize attempts made to modify AI output so it more closely resembles natural human language. This process uses algorithmic changes to tone, structure, and phrasing rather than actual human emotion or intent.

About forum moderation, you make a fair point. Finding AI-generated content can be difficult, particularly in cases when users fail to reveal their use of such instruments. Consequently, it makes sense that moderators give content quality and relevance top priority over the way the post was produced. Posts judged as spam, from anywhere, are more likely to be deleted.

GPTZero: 8% AI-generated
Sapling.ai: 0% Fake
Copyleaks: No AI Content Found
Quillbot: 100% AI-generated

Surprisingly, the detector from the same source that was used to humanize the text has detected that it's AI-generated, and all other tools refused to accept that. However, this is just an experiment done quickly with no edits or anything like that, but imagine if someone spends a few minutes on each post, running it through two or more such tools, refining the response a bit more in each attempt, they could come up with totally undetectable outcomes.

It's true, these tools are only good at fooling detectors, and a human, reading the post, should be able to determine that it's written and refined by AI, but then, you can't take actions based on your assumptions or even knowledge, and you need proof to do that.

[LoyceV]

Surprisingly, the detector from the same source that was used to humanize the text has detected that it's AI-generated

It probably stores everything it ever created, so in this case they know how it was generated. Now add some spelling mistakes and make the language less complicated, and it'll be harder to detect.

But:

Response generated through ChatGPT:

I'd say it's a generic shitpost. It uses fancy words and most of it is correct, but it doesn't add anything. It sounds like listening to the average politician: using as many words as possible to say nothing the reader doesn't know yet, but you can agree with it because it's (more or less) correct.