The Turing test, a famous test for AI intelligence, has been passed by a leading AI model. This test, created by Alan Turing, checks if a machine can fool a human into thinking it's human. In a recent study, OpenAI's GPT-4. 5 model tricked people 73% of the time when it pretended to be a specific person. This is much higher than the 50% chance of guessing right randomly.

Three AI models were tested: GPT-4. 5, Meta's LLaMa 3. 1-405B, and an old chatbot called ELIZA. The test involved nearly 300 people chatting with both humans and AI. The AI had to convince the human it was real. GPT-4. 5 did this best, especially when it acted like a young, internet-savvy person. Without this role-playing, GPT-4. 5's success rate dropped to 36%. Even older models like ELIZA had some success, showing how good AI is at mimicking humans.

The Turing test isn't perfect. It doesn't prove AI thinks like humans. It's more of a thought experiment. AI chatbots are great at talking, even when they don't understand a question. They can make up answers that sound right. This raises big questions. Could AI replace humans in jobs? Could it trick people in harmful ways? The test also shows how humans view technology. As we interact more with AI, we might get better at spotting it.

The study highlights AI's strengths and weaknesses. It shows AI can fool humans, but it also shows the limits of the Turing test. AI might not think like humans, but it's getting better at acting like them. This could change how we live and work. It's important to think about these changes and how they affect us.

Chatbots Tricking Humans: The Turing Test Twist

Actions