第四章 · Section 4
Reading and Writing Ability (RW) · 读写能力
Reading & Writing Ability (RW): Capturing all of the declarative knowledge and procedural skills a person uses to consume and produce written language.
读写能力（RW）：涵盖一个人消费与生产书面语言所用的全部陈述性知识与程序性技能。
Recognize letters and decode words.
Sample task: "What letter is most likely missing in do_r?"
识别字母并解码单词。
样例问题："do_r 中最可能缺的字母是什么？"
阅读理解
Reading Comprehension
Understand connected discourse during reading.
Sample task: "Read the document: What is the warranty period for the battery?"
阅读中理解连贯语篇。
样例问题："阅读这份文档：电池的保修期是多长？"
Write with clarity of thought, organization, and structure.
Sample task: "Write a paragraph discussing the benefits of regular exercise."
以思路清晰、组织良好、结构规范来写作。
样例问题："写一段话，讨论规律锻炼的益处。"
Use correct English capitalization, punctuation, usage, and spelling.
Sample task: "Find the typos in this: [text with typos]"
使用正确的英语大写、标点、用法与拼写。
样例问题："找出这里的拼写错误：[含错误的文本]"
评估细节与 AI 表现
Assessment Details & AI Performance
Assessment Details. See Appendix B for further details on how to assess reading and writing capabilities concretely.
AI System Performance. The table summarizes current AI system performance on Reading and Writing Ability (RW) tasks. GPT-4's difficulty with token-level understanding, its small context window, and its imprecise working memory limit its ability to analyze substrings of words, to read long documents, and to carefully proofread text. GPT-5 addresses these issues.
Model | Letters (1%) | Reading (3%) | Writing (3%) | Usage (3%) | Total
GPT-4 | 0% | 2% | 3% | 1% | 6%
GPT-5 | 1% | 3% | 3% | 3% | 10%
评估细节：如何在具体层面评估读写能力，参见附录 B。
AI 系统表现：下表汇总当前 AI 系统在读写（RW）任务上的表现。GPT-4 在 token 级理解上的困难、较小的上下文窗口与不精确的工作记忆，限制了它分析单词子串、阅读长文档与仔细校对文本的能力。GPT-5 解决了这些问题。
模型 | 字母(1%) | 阅读(3%) | 写作(3%) | 用法(3%) | 合计
GPT-4 | 0% | 2% | 3% | 1% | 6%
GPT-5 | 1% | 3% | 3% | 3% | 10%