Artificial intelligence (AI) is increasingly becoming a valuable tool in our daily lives, assisting with everything from answering questions and performing work tasks to aiding in studies. AI’s efficiency in completing trained tasks often surpasses human capabilities, and its adaptability means it can be trained for almost any purpose. This was certainly the case for an Argentinian man who trained an AI to assess if his future girlfriends might be similar to his ex-partner. He used Tinder matches, interacting with dozens of women, but found that nearly all displayed toxic traits, with only one passing his AI’s filter.
We’ve grown accustomed to having AI as a personal assistant, readily available for help. Some use it for basic queries instead of searching online, others present existential doubts to see if AI can offer solutions, and some leverage it to debug code. While these are diverse applications, AI’s potential extends further. By training it with specific datasets, it can acquire specialized knowledge and adapt to various other functions.
Training an AI to Detect Toxic Tinder Matches, Similar to His Ex
As explained by teKa on X, Marcos, a 31-year-old from Cordoba (presumably Argentina, based on his profile), ended his four-year relationship with his girlfriend, Carla, after enduring multiple infidelities. To make matters worse, his ex emptied his bank account before leaving, leaving him in a dire situation. Determined not to repeat such an experience, Marcos decided to train an AI to identify if other women might exploit him similarly. He utilized Anthropic’s Claude AI model and performed ‘fine-tuning’ using 38,000 WhatsApp messages, representing four years of chats with Carla.
This training allowed the AI to recognize specific expressions and behaviors, enabling it to analyze potential future disasters. Marcos discovered certain patterns that preceded negative outcomes and indicated his ex’s toxic tendencies. The next phase of his plan involved meeting new women and potential partners to run through his AI-created toxicity detector. Tinder was the obvious choice, offering a direct and efficient way to connect with people without unnecessary detours.
Out of 24 Tinder Matches Over Several Months, 23 Failed the Toxicity Test, and 3 Later Asked for Money
Every match Marcos received on Tinder was put through his AI toxicity detector, named Exfilter. Within seconds, the AI could analyze the Tinder profiles and chat messages of these women, identifying similarities to his past interactions. The results were astonishing: over four months, he disqualified all 23 women who matched with him. Even more concerning, months later, three of these disqualified women asked him for money.
Ultimately, only the 24th match passed the test. She achieved a ‘toxicity score’ of only 12 out of 100, while the others failed and exhibited similarities to his ex based on the AI’s analysis. Exfilter identifies five key ‘red flags’:
- Early love bombing
- Passive victimhood
- Triangulation (repeatedly mentioning friends)
- Ambiguity (canceling plans with little notice)
- Financial red flags (mentioning travel and high spending despite no apparent work or income)
Update: There is uncertainty regarding the veracity of this story, as comments suggest that the Exfilter GitHub repository does not exist, nor are there any ‘fine-tuned’ Claude models available.
SEO Title: Argentinian Man Uses AI to Filter Toxic Tinder Matches – Only 1 Out of 24 Approved
SEO Description: Discover how a man trained an AI on years of text messages to detect toxic behaviors in potential partners on Tinder. Find out how many matches passed his AI’s rigorous ‘red flag’ test.
