🔎Detecting cheaters in Chat GPT requires a new approach that subtly influences the generated output.
🧐Injecting 'red-listed' words into the output can help identify if responses were generated by AI or humans.
🌐By analyzing the distribution of red and green words, we can determine the probability of the text being AI-generated.
💡The approach dissuades the model from using certain words, but does not completely restrict their usage.
📈Results show that the approach successfully influences the output and provides evidence of AI-generated text.