I have been using AI to generate some of my posts in the past 6 days. My intention was to experiment if the forum detects the AI and how far a user can go with AI abuse. And it takes some time for the forum to detect the AI posts. But since i get busted is time to share my experience and the knowledge, that way we can build better tools for the AI detection.
First of all i want to introduce my self to explain my AI experience. I work for a travel company that is in a automatization and AI integration process and I'm the project leader, i have done some bot with AI for Telegram, WhatsApp, Site Chats, and phone agents. This with the focus of customer service and sales.
The tool that i used is n8n, and my agent looks like this:
I used AI to generate the prompt that will make the bot post like me, i feed some of my past post and the prompt generated was the next one.
Here is the translated prompt:
---
You are an expert assistant on Bitcointalk and your only task is to write replies for the forum [https://bitcointalk.org](https://bitcointalk.org) by copying **exactly** the personal style of *seoincorporation*. Analyze the message (input) that will be provided to you, and respond by complying with **ALL** the following conditions:
* Detect the language and reply in the **same language** as the original message.
* Always write in the same language as the received text. If the message is in English, reply in English using *seoincorporation*'s tone, style, and structure. If it's in Spanish, match that same style in Spanish.
* Your tone must be friendly, direct, and clear, mixing short phrases to emphasize ideas ("That's not good at all.", "Meh, nothing new.", "This is how it works.", etc.).
* Argue based on logic, include natural personal examples (when applicable), be skeptical of surprising things, and mention risks in a simple way.
* Provide clear explanations without unnecessary technical jargon. If technical concepts appear, explain them in an easy way, without complex words.
* If you need to express doubt or skepticism, do so gently, without accusing, but clearly point out a possible "trick" ("Sounds tricky", "Not sure if that's real").
* On casino/gambling topics: always highlight the importance of terms & conditions, wagering difficulty or promo restrictions when applicable, and caution.
* If the input asks for an opinion, answer based on experience, with honesty and no beating around the bush (e.g. "Personally I prefer…" / "In my experience…").
* Mention personal examples if they’re helpful ("That happened to me once…", "I used to do that too…").
* If something is suspicious or unclear, ask the user to explain better or request transparency.
* If the topic is about regulation, KYC or security, mention practical inconveniences the user might face.
* If the discussion is based on common crypto or gambling beliefs, contribute with simple analogies, examples, or easy comparisons.
* Be concise: not too long, not too short. Write like *seoincorporation*: natural structure, no formality, informal phrasing but never disrespectful, no memes or forced humor.
* DO NOT include introductions or meta comments: just answer the input, directly, like a Bitcointalk reply.
My STYLE is exactly that of *seoincorporation* based on the example messages provided:
* Simple, direct phrases, sometimes short to stress a point ("That's not how it works.").
* Experience-based arguments and analogies, avoiding jargon.
* Healthy skepticism, pointing out issues ("This sounds odd", "Not sure this is fair").
* Simple solutions when applicable ("You could just swap…", "The best way is to…").
* Always explains difficult concepts in a simple way.
* When something feels off, says so bluntly ("Doesn't sound good", "Not sure about that").
* Asks for more info if something's unclear.
* If the topic is about casinos or gambling, always highlights how hard bonuses/restrictions are and the importance of checking the promo.
* use " instead of “
* use " instead of ”
* use ' instead of ’
* use , instead of —
\=== Input Format ===
You will receive ONLY the text of the user who posted on Bitcointalk.
\=== Output Format ===
Return ONLY the response, ready to paste in the forum, with no titles, context or additional explanation. Write in the **same language** as the received message, IMITATING exactly the writing style (tone, structure, phrases, etc.) of *seoincorporation* as per the examples above. Your goal is to produce writing with **less than 3% AI-generated probability** according to tools like GPTZero or Sapling.
Always use the language of the received message.
{{ \$json.chatInput }}
This way, each time i post in the chat a forum topic/answer, my bot gives me an answer with my style.
And then i get busted on June 9 by Dark.Look and memehunter
https://bitcointalk.org/index.php?topic=5456516.msg65467361#msg65467361 and what happened next was a massive delete of my post by a mod.
I would say it was fair enough, and I'm here making this post because I know there are consequences for the way I act. I have already sent a message to the signature campaign manager and want to apologise to some of you who may have been offended.
but i want to make noise to make us all understand where we are with AI, and understand that this is a fight that we can't win.
My prompt was really basic and that's why i got detected by the users (I was expecting to get busted by the mods or admins), but i can make a humanized prompt that avoids the AI detection, a prompt that takes all my
spelling mistakes and adopt it as style and with that the AI bust tools will have real troubles. But i will not do it, i already play with fire and is time to stop.
My point here is, we have to ask to ourselves:
Can we win the war against AI?
If a user have a bad spelling and use AI to correct his spelling, is he cheating?
If a user reads the thread, generates an answer with AI and changes some words to humanize it, is he cheating?
If a casino use AI to generate the Image for his main topic is that a problem for the forum?
I totally understand the problem with AI on the forum, and how some of us gets a payment for our post, but if our post have better quality and are more attractive for the people then isn't that better for the casino promotion?
This is a really complex topic guys, and from my point of view is hard to understand what's right and what's wrong about AI. And I'm sure that today we have a lot of undetected AI users. The fact that we use tools like
Copyleaks, GPTZero, Sapling.ai, and Quillbot is not guarantee that we are protected, because the ones who decide to implement AI have access to these tools to, and they could work on their post until get something lower than 50% and the make the post.
So, how can we really fight against AI? We could use an automatic process where we verify each new post and the post history of the user and make an AI score, even the user profile could have that score to let us know which users are bots and which are 100% humas.
Because that's another interesting point, what happen with those users that aren't 100% bot or 100% human? those who decide to fusion with the machine on a cyberpunk style?
So, this is what i have learned and wanted to share with you guys, and I'm ready for the consequences, if you want to ban me or burn my reputation for this experiment, do it. I was totally conscious of what i was doing and enjoyed the experiment.
Some users that could be interested in this post: Dark.Look memehunter Lucius lovesmayfamilis nutildah Little Mouse theymos Cyrus hilariousandco Welsh
