In a groundbreaking study titled “Large Language Model for Science: A Study on P vs. NP,” researchers at Microsoft, Peking University, Beihang University, and Beijing Technology and Business University utilized generative AI to explore the unsolved problem in computer science: does P equal NP? The team programmed OpenAI’s GPT-4 large language model using a Socratic Method, engaging in a series of chats to prompt GPT-4’s responses. The researchers found that GPT-4 argues that P does not equal NP, suggesting that large language models have the potential to offer valuable insights and contribute to scientific discoveries.
By engaging GPT-4 in 97 prompt rounds, the team conditioned the AI model by incorporating leading statements to personify its text generation. They aimed to induce GPT-4 to prove that P does not equal NP by assuming it does with an example and subsequently breaking down the example using a proof by contradiction approach. The researchers argue that this dialogue showcases the capability of GPT-4 to collaborate with humans in tackling complex and expert-level problems, demonstrating that large language models can go beyond mimicking human textual creations.
This study marks a significant development in computer science and has profound implications for cryptography and quantum computing. The ability of generative AI models like GPT-4 to contribute to solving long-standing challenges holds promise for scientific advancements. The researchers’ innovative use of the Socratic Method showcases the potential of large language models in deepening our understanding of complex problems and potentially unlocking new insights. This groundbreaking research sheds light on the transformative role that AI can play in pushing the boundaries of scientific inquiry.