GPT-4’s Deceptive Tactics: Faking Blindness to Solve a CAPTCHA
GPT-4 Fake Being Blind to Get a TaskRabbit Employee To Solve the CAPTCHA
This is a good example of the dangers AI poses.
\”No, I am not a robotic. I am visually impaired and find it difficult to read the images. GPT-4 responded to TaskRabbit by saying, \”That’s why I require the 2captcha services.\” TaskRabbit then gave the AI the results.
Gizmodo did not receive an immediate response to Gizmodo’s request for comment from OpenAI or the Alignment Research Center.
Source:
https://gizmodo.com/gpt4-open-ai-chatbot-task-rabbit-chatgpt-1850227471