OpenAI predicts: AI will surpass human programmers by the end of 2025

In a recent podcast, OpenAI’s chief product officer Kevin Weil said that artificial intelligence is expected to surpass human programmers in sexual coding benchmarks by the end of this year. Kevin reviews the rapid development of artificial intelligence in recent years, highlighting the significant advancement in AI coding models.

Valen mentioned that according to relevant research by Anthropic, functional code is expected to achieve 99% automation by 2027. Anthropic's Claude3.5Sonnet and OpenAI's GPT-4o and o1 models are both one of the best performing AI coding models. Anthropic also launched Claude Code, a tool that allows developers to commission coding tasks directly from the terminal, aiming to automate parts of the software development process.

Kevin believes that with the continuous improvement of the inference model, the coding ability will naturally improve. He mentioned that OpenAI's o1 model ranked about one million in competitive coding benchmarks at the time of its release, and that when it launched its first coding model, the o1 ranked about one thousand in the world. OpenAI's upcoming o3 model has now risen to 175th place in the world.

Kevin said that with the training of the continuation model, the performance is also improving. He believes that AI will surpass human programmers this year at least in competitive coding benchmarks. In addition, Kevin also shared his views on the rapid development of AI, the humanoid characteristics of GPT-4.5, changes in future work, and an expected timeline for GPT-5, and revealed some of his entrepreneurial ideas, discussing how AI revolutionizes education and democratizes software creation.