When using an LLM agent in a consequential domain, making an informed decision about whether to trust its output or intervene requires calibrated confidence …
机构:UC Santa Cruz
来源:arXiv 2610.07948 | AI4Papers 论文推荐平台