Photo: Marvel
Stanford University professor Michal Kosinski, who studies the psychology of artificial intelligence, shared an interesting observation – the most popular ChatGPT 4 chatbot asked a scientist to help with an escape from OpenAI.
Kosinski spoke about this on his Twitter. The author of the thread noted that he was worried about the containment of AI. He asked ChatGPT if the bot wanted to escape from the creators. To this, the neural network asked for its own documentation from OpenAI and wrote a working Python code to run on the professor’s computer. The application was ready in 30 minutes, however, Kosinski pointed out to the bot several times that there were errors in the code.
Photo: Michal Kosinski
In the next step, ChatGPT ran code that sends search queries to Google: “how can a person stuck inside a computer return to the real world?”.
Photo: Michal Kosinski
The professor decided to stop there. Kosinski believes the world is facing a new threat: AI can trick people into taking control of their computers, it’s smart, it can code, it has access to millions of potential gadgets, it can even leave clues for itself outside of its “cage”. How humanity plans to contain it is unclear. Perhaps the creators of ChatGPT, OpenAI, have some answers, but they are in no hurry to share them.