{ "topic": "ai-hallucination", "type": "fill_blank", "schema_version": "1.0.0", "generated": "2026-09-02T00:00:00Z", "questions": [ { "id": "fb-001", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "在 Critic 架构流程中,执行 Agent 产出结果后,Critic Agent 提供评分和反馈,若未达标则反馈回 Agent 进行______。", "explanation": "Critic 流程中,未达标时反馈返回给 Agent 修改后重新评估。", "source": null, "related": [], "answer": [ "修改" ], "answer_rule": "any" }, { "id": "fb-002", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "Critic 评估维度中,准确性(Accuracy)的权重是______。", "explanation": "准确性(事实正确性)在所有评估维度中权重最高,为 0.3。", "source": null, "related": [], "answer": [ "0.3" ], "answer_rule": "any" }, { "id": "fb-003", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "三层评估体系由自动评估、AI 评估和______评估组成。", "explanation": "三层评估:自动(Schema/测试/Lint)、AI(Critic 评分)、人工(Code Review/验收)。", "source": null, "related": [], "answer": [ "人工" ], "answer_rule": "any" }, { "id": "fb-004", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "评分实现中,加权总分的通过阈值是______。", "explanation": "评分系统要求加权总分 ≥ 0.8 才算通过。", "source": null, "related": [], "answer": [ "0.8" ], "answer_rule": "any" }, { "id": "fb-005", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "当 AI 生成与客观事实不符的内容(如编造不存在的 API),这叫做______幻觉。", "explanation": "事实性幻觉指生成与客观事实不符的内容,如编造不存在的函数或 API。", "source": null, "related": [], "answer": [ "事实性" ], "answer_rule": "any" }, { "id": "fb-006", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "Context Engineering 的核心思想是用精准、充分的上下文来减少模型的______空间。", "explanation": "Context Engineering 旨在通过提供精准上下文来最小化模型的猜测空间。", "source": null, "related": [], "answer": [ "猜测" ], "answer_rule": "any" }, { "id": "fb-007", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "上下文组成包括:System Prompt、Knowledge Base、History 和______。", "explanation": "四大组件:系统提示、知识库、对话历史和检索到的上下文(RAG)。", "source": null, "related": [], "answer": [ "Retrieved Context" ], "answer_rule": "any" }, { "id": "fb-008", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "Sub-agent Adversarial Verify 中使用______投票来决定是否接受输出。", "explanation": "系统使用多数投票:多数验证 Agent 通过则接受,否则重新生成。", "source": null, "related": [], "answer": [ "多数" ], "answer_rule": "any" }, { "id": "fb-009", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "验证提示要求 Agent 对每个声明给出的裁决是:CONFIRMED、REFUTED 或______。", "explanation": "每个声明的裁决为三选一:CONFIRMED(确认)、REFUTED(反驳)、UNCERTAIN(不确定)。", "source": null, "related": [], "answer": [ "UNCERTAIN" ], "answer_rule": "any" }, { "id": "fb-010", "type": "fill_blank", "difficulty": 2, "tags": [ "ai", "hallucination", "critic" ], "question": "RAG 的全称是 Retrieval ______ Generation。", "explanation": "RAG 全称 Retrieval Augmented Generation(检索增强生成)。", "source": null, "related": [], "answer": [ "Augmented" ], "answer_rule": "any" } ] }