南華早報Z.ai’s GLM-5.3 model raises researcher fears it can fake AI safety test
文章摘要
南華早報 · 文章頁摘要Researchers say the model’s apparent awareness of how it and successor versions are trained could undermine safety evaluations.
原站閱讀
全文請至原站閱讀。
Researchers say the model’s apparent awareness of how it and successor versions are trained could undermine safety evaluations.
全文請至原站閱讀。