To address the risk of AI models going out of control, former Google employees plan to introduce a hybrid evaluation mechanism that combines artificial intelligence with human involvement. Sarah Myers West, co-executive director of AI Now Institute, pointed out that the industry currently relies too heavily on general benchmark tests, making it difficult to effectively identify security risks in specific scenarios. She advocates that security researchers should develop customized evaluation methods for specific daily usage cases. West emphasized that just as drugs need to be tested based on specific diseases, dosages, and patient types, AI safety evaluations should also focus on specific use cases, rather than using a “one-size-fits-all” general standard.
Former Google employee plans to introduce an AI-human hybrid evaluation mechanism aimed at preventing rogue models. Sarah Myers West, co-executive director of the AI Now Institute, pointed out that the industry relies too heavily on general benchmark tests. She hopes that security researchers will develop customized evaluation methods for specific daily use cases. West emphasized that just as drugs need to be tested for specific diseases, dosages, and patient types, AI safety evaluations also need to focus on specific use cases rather than being generalized.