Tenzai, a cybersecurity company, won a three-month hacking contest by using Chinese AI systems from Z.ai rather than U.S. models like Claude or GPT. The victory demonstrates the power of China's 'open weight' AI approach, where models are freely available online. U.S. companies like Anthropic and Open AI tightly control their systems, while Chinese firms release theirs without restrictions. This has raised concerns among U.S. experts about potential misuse by malicious actors.
However, many in the AI community argue open-weight systems are safer because they enable both offensive and defensive capabilities. 'With open-weight systems, we can increase our defensive capabilities by 100 times,' said Johnny Wang, former Google security researcher and CEO of Hinoki. The gap between Chinese and U.S. AI performance is narrowing, with Chinese models like GLM from Z.ai now matching leading U.S. technologies in some areas. 'You can't put the genie back in the bottle,' said Pavel Gurvich, Tenzai's CEO.
Source: New York Times Top Stories · Summarized by HeadlinesBriefing