OpenAI proposes new technology PVG to solve the "black box" problem

On July 18th, according to the Economic Observation Network, OpenAI's official website said today that with the use of ChatGPT in the fields of finance, law, and marketing, there will be a "black box" situation due to the inability to verify its accuracy. In order to solve this problem, OpenAI has proposed a new training framework, Prover-Verifier Games (referred to as "PVG"). For example, using small models such as GPT-3 to verify and supervise the output of GPT-4 large models, thereby improving output accuracy and controllability. Recent news shows that OpenAI is making the ability of artificial intelligence (AI) large models closer to the level of human-like reasoning.