45 Matching Annotations
  1. May 2026
    1. Rather than supervising what the agent does, we supervise what it's _able_ to do by enforcing access boundaries through, for example, sandboxes, virtual machines, and egress controls.

      行动建议:为AI代理系统实施环境层边界控制,使用沙盒、虚拟机和出口控制技术限制代理的访问能力,而不是仅仅依赖行为监督。这种方法能够从根本上限制代理可能造成的损害范围,即使模型层防御失效。

  2. Apr 2025
  3. Feb 2025
  4. Apr 2022
    1. Youyang Gu. (2021, May 25). Is containing COVID-19 a requirement for preserving the economy? My analysis suggests: Probably not. In the US, there is no correlation between Covid deaths & changes in unemployment rates. However, blue states are much more likely to have higher increases in unemployment. 🧵 https://t.co/JrikBtawEb [Tweet]. @youyanggu. https://twitter.com/youyanggu/status/1397230156301930497

  5. Oct 2021
  6. Jul 2021
  7. Jun 2021
    1. Christophe Fraser 💙 on Twitter: “Reading Cummings accounts of early creation of Test & Trace, a question I have is when and how it was morphed from aiming to find ~30 contacts per index case, needed to contain spread, into a service that contacts 2-4 contacts per index case, mostly within household.” / Twitter. (n.d.). Retrieved June 28, 2021, from https://twitter.com/ChristoPhraser/status/1408454903249477632

  8. May 2021
  9. Jan 2021
  10. Dec 2020
  11. Aug 2020
  12. Jun 2020
  13. May 2020
  14. Apr 2020
  15. Dec 2018
  16. Nov 2015