Current alignment training cannot be expected to deliver what it is meant to deliver: AI agents that comply with norms whatever the gains and whether or not …
机构:DFKI
来源:arXiv 2611.02162 | AI4Papers 论文推荐平台