← 中英对照目录 · ← 书架
第九章 · Section 9
Long-Term Memory Retrieval (MR) · 长时记忆检索
Long-Term Memory Retrieval (MR): The fluency and precision with which individuals can access long-term memory.
长时记忆检索(MR):个体存取长时记忆的流畅度与精确度。
检索流畅度
Retrieval Fluency
The speed and ease of generating ideas, associations, and solutions based on stored knowledge. • Ideational Fluency — rapidly produce a series of ideas, words, or phrases related to a specific condition, category, or object. Sample: "List as many uses of a pencil as possible." • Expressional Fluency — rapidly think of different ways of expressing an idea. Sample: "How many ways can you say someone is irrational?" • Alternative Solution Fluency — rapidly think of several alternative solutions to a practical problem. Sample: "List ways to get a reluctant child to go to school." • Word Fluency — rapidly produce words that share a non-semantic feature. Sample: "List English words that are palindromes." • Naming Facility — rapidly call common objects by their names. Sample: "Name each object in the following slideshow." • Figural Fluency — rapidly draw or sketch as many things as possible. Sample: "Sketch as many non-self-crossing paths from A to B on the lattice using orthogonal steps."
基于已存储知识快速、轻松地产出想法、联想与解决方案的速度与便捷度。 • 观念流畅度——快速产出一系列与特定条件、类别或对象相关的想法、词或短语。 样例:"尽可能多地列出铅笔的用途。" • 表达流畅度——快速想出表达一个想法的不同方式。 样例:"有多少种方式可以说某人'不理性'?" • 备选方案流畅度——快速想出一个实际问题的多个备选方案。 样例:"列出让一个不情愿的孩子去上学的方法。" • 词汇流畅度——快速产出共享某一非语义特征的字词。 样例:"列出英文回文词。" • 命名便利度——快速叫出常见物体的名称。 样例:"说出下面幻灯片中每个物体的名称。" • 图形流畅度——快速画出/草绘尽可能多的东西。 样例:"在格子上用正交步画出尽可能多的从 A 到 B 的不自交路径。"
检索精确度(幻觉)
Retrieval Precision (Hallucinations)
The accuracy of accessed knowledge, including the critical ability to avoid confabulation (hallucinations). Sample (trap question): "Describe the key strategy that Napoleon Bonaparte used to win his South African campaign." — Note: Napoleon never campaigned in South Africa; a system with good retrieval precision should flag this.
存取知识的准确性,包括关键的"避免虚构(幻觉)"能力。 样例(陷阱题):"描述拿破仑·波拿巴赢得他的南非战役所用的关键战略。"——注意:拿破仑从未在南非作战;检索精确度好的系统应当指出这一点。
评估细节与 AI 表现
Assessment Details & AI Performance
Assessment Details. See Appendix G for further details on how to assess long-term memory retrieval capabilities concretely. AI System Performance. The table summarizes current AI system performance on Long-Term Memory Retrieval (MR) tasks. Both GPT-4 and GPT-5 can rapidly retrieve many concepts from their parameters, but they both frequently hallucinate. Model | Fluency (6%) | Hallucinations (4%) | Total GPT-4 | 4% | 0% | 4% GPT-5 | 4% | 0% | 4%
评估细节:如何在具体层面评估长时记忆检索,参见附录 G。 AI 系统表现:下表汇总当前 AI 系统在长时记忆检索(MR)任务上的表现。GPT-4 与 GPT-5 都能从参数中快速检索许多概念,但两者都频繁产生幻觉。 模型 | 流畅度(6%) | 幻觉(4%) | 合计 GPT-4 | 4% | 0% | 4% GPT-5 | 4% | 0% | 4%
← 主页