重點導讀
只公開 AI 模型還不夠,大家也需要知道它可能帶來哪些安全風險。
- 騰訊混元在公開 Hy4 權重的同時,也發布 System Card,分享安全、資安與對齊評估,並進一步檢視網路攻擊和生物科技方面的前沿風險。
- 評估還包括騰訊的玄武、朱雀與 Keen Labs,以及 FangcunLeap 和 Alice 的獨立安全測試。
- 公開這些檢查結果,能讓其他人瞭解模型受過哪些測試,並參考這些做法打造更安全的 AI。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
Tencent Hy@TencentHunyuan
·2 天前
Open weights. Built on trust. 🔓
Sharing how we test AI helps everyone build safer models.
Alongside Hy4's open weights, we're publishing our System Card, sharing our safety, security & alignment evaluations, with a closer look at frontier risks in cyber & bio.
This includes further reviews by Tencent's Xuanwu, Zhuque & Keen Labs, as well as independent safety & security testing by @FangcunLeap and @alice_dot_io.
Among the First Raising the bar for responsible AI in Asia. 🛡
System Card: https://t.co/VcFbN8xdu3
推文來源 · X (Twitter)在 X 參與討論
來源: Tencent Hy (x.com)
在 X 查看推文