arXiv cs.CY· Li Gan·· 9 天前AI 评分22
风险上限与开发截止日期：安全生产力不确定下如何为 AI 定节奏
Risk Ceilings and Development Deadlines: Pacing AI under Uncertain Safety Productivity
AI 导读
一项研究探讨监管者能否在安全生产力未知时同时承诺风险上限与开发截止日期：若最终模型未受保护的风险高于上限，就必须具备最低限度的安全知识储备，因此两项承诺只有在排除弱研究的前提下才成立。在仅含状态风险、安全产出恒定且知识完全迁移的校准设定下，先学习再重放开发、并把多余算力留给安全，所需生产力仅比必要边界高 3.3%。固定算力分配的规则锁定的是日期而非风险。
正文 · 原文
Abstract:Can a regulator promise both a risk ceiling and a development deadline when safety productivity is unknown? A ceiling below the final model's unprotected hazard requires a minimum stock of safety knowledge, so both promises hold only if weak research can be ruled out. Learning first and then replaying development, with spare compute in safety, comes close to that minimum. In a calibration with only state risk, constant safety yield, and full knowledge transfer, it needs only 3.3 percent more productivity than the necessary bound. Customer services pay for the guarantee. Rules that fix compute allocation fix dates, not risk.
| Subjects: | General Economics (econ.GN); Computers and Society (cs.CY) |
| Cite as: | arXiv:2610.11093 [econ.GN] |
| (or arXiv:2610.11093v1 [econ.GN] for this version) | |
| https://doi.org/10.48550/arXiv.2610.11093 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Li Gan [view email]
[v1]
Thu, 8 Oct 2026 02:10:00 UTC (59 KB)
来源：arXiv cs.CY · arxiv.org