OpenAI researcher Yao Shunyu: AI will move from solving problems to defining them

A few days ago, Yao Shunyu, who graduated from the Yao class at Tsinghua University and is now at the OpenAI Research Institute, released a blog post discussing his predictions for the future development of AI. Shunyu Yao reviewed the development history of AI. For decades, he said, AI has focused on developing new training methods and models, and has achieved remarkable successes, all of which stemmed from fundamental innovations such as search, deep reinforcement learning (Deep RL), and reasoning. And now that Deep Reinforcement Learning is finally starting to generalize, the landscape of AI empowerment for humans has changed. According to Shunyu Yao, with the breakthroughs in reinforcement learning, AI is beginning to solve diverse tasks such as software engineering, creative writing, and IMO-level math problems. With the introduction of language and reasoning, AI is able to generalize tasks across domains and solve complex problems. Yao Shunyu also mentioned that the second half of AI will shift from solving problems to defining them, and innovation in evaluation methods will be key.