OpenAI CEO警告：美国严重低估中国AI实力，出口管制也没用

Group 1 - OpenAI CEO Sam Altman warns that the U.S. may underestimate the complexity and severity of China's advancements in artificial intelligence (AI) and suggests that relying solely on export controls may not be a reliable solution [1] - Altman expresses concerns about the intertwined nature of the U.S.-China AI competition, indicating that the implications extend beyond a simple "who is leading" narrative [1] - Despite the U.S. escalating semiconductor export controls, Altman doubts the effectiveness of these policies in keeping pace with technological advancements [1] Group 2 - Altman notes that competition from Chinese open-source models, such as DeepSeek, has influenced OpenAI's decision to release its own open-weight models, marking a significant strategic shift for the company [3] - OpenAI recently released two open-weight language models, gpt-oss-120b and gpt-oss-20b, which are designed for developers, researchers, and companies to download and customize locally [3][4] - The release of open-weight models provides transparency and control but does not equate to open-source, as OpenAI has not disclosed its training data or complete source code [4] Group 3 - The external reception of OpenAI's open-weight model release is mixed, with some developers criticizing the models for lacking the powerful features found in OpenAI's commercial products [4] - Altman acknowledges that the team intentionally optimized the models for a core application scenario: local programming agents, while also allowing for potential adaptation to other uses if demand changes [5]