AIGC (Artificial Intelligence Generative Content) refers to the creation and generation of various types of content using artificial intelligence (especially AI large language models). These contents include, but are not limited to, text, image, audio, video, and code.
AIGC（Artificial Intelligence Generative Content，人工智能生成内容) 指的是利用人工智能技术(特别是AI 大语言模型)创建和生成各种类型的内容。这些内容包括但不限于文本、图像、音频、视频和代码。
AIGC (Artificial Intelligence-Generated Content) application delivered by higher education workers can enrich contents and teaching efficiency.
高等教育工作者应用 AIGC(人工智能生成内容) 可以丰富教学内容并且提高教学效率。
This AIGC WORKSHOP are divided into three stages:
1) Prompt engineering and Prompt construction
2) Talk to AIGC
3) Parse document using AIGC.
Educators can use it to understand common AIGC tools. Content generated by AIGC tool is obtained by constructing high- quality prompts, and the content generated by AIGC tool is used to assist instructional design.
本AIGC工作坊将分将分三个阶段进行:
1) 提示工程和提示词构建
2) 与AIGC对话
3) 使用AIGC解析文档
教育工作者可透过本工作坊快速了解常用AIGC工具的使用。通过构建高质量的提示语来获取有价值的AIGC工具生成内容, 借助AIGC工具生成内容来辅助教学设计。
The latest AI large language models are great for writing. How to use Prompt words correctly is key to obtain high-quality responses. In simple terms, it is an instruction called a Prompt, which is important in the AI large language model.
当前人工智能领域中的大语言模型在文本生成方面具有显著的优势。正确使用提示词是获得高质量回应的关键。简单来说，它是一种称为提示的指令，提示词是一种指令，它在人工智能大语言模型的操作中非常重要。
For example: If AI large language model equals to an employee, a Prompt is equivalent to a specific instruction given to the employee. The detailed instructions determine the output effect. High-quality Prompt words can significantly improve AIGC content relevance and accuracy.
例如：如果将人工智能大语言模型比作一名员工，而提示词则相当于向该员工发出的具体工作指令。这些详尽的指令直接影响着最终的工作成果。高质量的提示词能显著提升人工智能生成内容的相关性和准确性。
Structural requirements expression can improve the quality of output responses. A logical prompt should specify:
1. Role: Role, such as “Writer”, “Designer”, or “Historian”.
2. Task: Task, such as “Write a novel” or “Design a logo”.
3. Format: Format, such as “Markdown format” or “PDF format”
结构化地表达需求,可以提高输出回答的质量。一个合乎逻辑的提示词应当涵盖以下要素：
1. 角色：指导模型扮演一个特定角色，比如“作家”、“设计师”或“历史学家”等。
2. 任务：向模型明确指出你希望其完成的具体任务，比如“写一篇小说”或“设计一个标志”等。
3. 格式：指定期望的输出格式，如“Markdown 格式”或“PDF 格式”。
For example:
Role: You are Marx.
Task: You need to give some advice to Chinese college freshmen at the beginning of the school year. The advice should be in-depth and practical.
Format: No more than 200 words, please express in English
例如:
角色：你是马克思。
任务：你需要给中国的大学新生一些开学建议，要求有深度且可落地。
格式：字数不超过 200 字，请以英文方式表达。
This workshop gives eight advanced Prompt word frames. When choosing the Prompt word frame, it is necessary to determine the appropriate frame according to specific application scenario and the parts of framework are to be clearly defined. 本工作坊提供八个进阶提示词框架, 选择提示词框架时，需要根据具体的应用场景和目标来确定最合适的框架。针对选择的框架, 需要对框架的各部分进行清晰的定义。
For example: A RACE framework is suitable for tasks where roles, actions, contexts, and expectations are specified, in education, training, and role-playing activities.
Choose the RACE framework to clearly define participants, roles, actions, contexts, outcomes.
例如: RACE框架适用于需要明确角色、行动、背景以及期望的任务，特别是在教育、培训和角色扮演类的活动中。
需要清晰地定义参与者的角色、他们要采取的行动、所处的上下文以及期望的结果时，选择RACE 框架。
- Specify a role for the model - 为模型指定一个角色
- Use separators to clarify different inputs - 使用分隔符明确输入的不同信息部分
- Write coherent instructions - 编写清晰的说明
- Define steps to complete task - 指定完成任务所需的步骤，1. 2. 3...
- Provide examples sample - 提供示例样本，例如：...
- Format and output length - 指定格式及所需的输出长度
- Provide reference text - 提供参考文本
- Assist AI model answer with less fiction - 为AI大语言模型提供参考文本可以帮助它以较少的虚构进行回答
- Use intent classification to identify relevant Prompt for user queries - 使用意图分类来识别用户查询相关的Prompt
- Dissect complex tasks into simple subtasks - 将复杂任务分解为简单子任务
- Summarize or filter previous dialogues for lengthy conversations - 对于长对话，总结或过滤以前的对话
- Summarize documents - 对文档进行总结
- Advise AI language model for solutions - 指导AI大语言模型自行找出解决方案
- Give sufficient time for model to "think" - 给予模型足够的时间进行"思考"
- Use phrases like "think step by step" - 使用”一步一步思考“类似的话术
- Ask for anything is missing - 询问是否遗漏了任何内容
- First diverge , then summarize - 先发散后总结
- Generate multiple answers, ask AI model reasons for various answers, and synthesize - 先生成多个答案，询问AI大模型给出多种答案的理由，最后综合
提示:
文心一言 (https://yiyan.baidu.com/)
Kimi (https://kimi.moonshot.cn/)
讯飞星火 (https://xinghuo.xfyun.cn/desk)
OpenAI cookbook: https://cookbook.openai.com/
This section is required verify that our prompt words are valid on AIGC tool.在这一部分,需要在AIGC工具上验证提示词是否有效。
It is noted that AIGC tools (such as chatGPT) cannot be used due to restrictions in China. This workshop uses three local popular AIGC tools instead. Click links at the beginning of the following paragraphs to access:由于中国大陆的相关限制，我们无法直接访问一些来自国外的AIGC工具（例如：ChatGPT)，因此本工作坊选择了三款在中国大陆流行的AIGC工具，点击文段开头的链接即可访问对应的AIGC工具链接, 三个AIGC工具分别是:
1.Wenxin Yiyan(https://yiyan.baidu.com/) is a large-scale language model developed by Baidu for artificial intelligence, to predict and generate the next sentence based on previous one. Users can interact with Wenxin Yiyan through dialogue, ask questions, or requests by commands input to obtain information, knowledge, and inspiration.
文心一言 (https://yiyan.baidu.com/) 是百度研发的人工智能大语言模型产品，能够通过上一句话，预测生成下一段话。 用户可以通过输入【指令】和文心一言进行对话互动、提出问题或要求，让文心一言高效地帮助人们获取信息、知识和灵感。
2. Xunfei Spark (https://xinghuo.xfyun.cn/desk) is a new generation of cognitive intelligence models launched by iFLYTEK. It has cross-domain knowledge and language comprehension capabilities to integrate multimodal input and output. It can also perform tasks based on natural dialogue, evolve on massive data and large-scale knowledge to achieve a full process closed loop from proposal to problem-solving.
讯飞星火 (https://xinghuo.xfyun.cn/desk) 是科大讯飞推出的新一代认知智能大模型，具有跨领域的知识和语言理解能力，融合多模态输入和输出，能够基于自然对话方式理解与执行任务，从海量数据和大规模知识中持续进化，实现从提出、规划到解决问题的全流程闭环。
3. Kimi (https://kimi.moonshot.cn/) is a large-scale AI model developed by Beijing Moonshot AI Technology Co., Ltd. It supports text input and is a unique tool for lengthy text processing without segmentation and has a powerful memory function to maintain conversation integrity and coherence.
Kimi (https://kimi.moonshot.cn/) 是由北京月之暗面科技有限公司（Moonshot AI）开发的智能助手。支持超长文本输入，是全球范围内罕见的超长文本处理工具，用户无需分段处理资料。拥有强大记忆功能， 可以长时间保持对话完整性和连贯性。
The above three figures show the results of applying Prompt words designed in the first part for each of these three AIGC tools to output corresponding AIGC content. 以上三张图展示了我们应用第一部分中设计的提示词分别在三个AIGC工具中进行测试的结果。可以看到三个工具都输出了相应的AIGC内 容。
Timeliness: Prompt Prompts Knowledge to interact with AI large language models comes from data uncovered without any information. For Example, if an user enquires of a recent scientific study, the model cannot provide information due to its knowledge cutoff date occurs usually before its training data cutoff date.
时效性: Prompt 提示词与 AI 大语言模型交互的知识来自训练时接触到的数据，无法提供未知信息或新的发现。例如：如果您询问关于一项最新科学研究的问题，模型可能无法提供相关信息，因为它的知识截止日期通常在训练数据截止日期之前。
Context Memory problem: The is caused by the byte length limit or some of its previous context information are lost when the model processes lengthy text. 上下文记忆问题: 上下文记忆问题是由于模型在处理超长文本时遇到了字节长度限制或者信息丢失的情况导致的。
AI illusion: When an AI linguistic model enquires of a fictional phenomenon, the response neither indicates its existence nor provide relevant information, but to emerge a plausible answer with references (as utter nonsense). AI幻觉:当人工智能语言模型询问一个虚构的现象时，模型给出的回答既不表明其存在也不提供相关信息，而是给出一个看似合理并且可以引用的答案（作为无稽之谈）。
Exact number of words: If a user wants to write a novel with a large language model with 1000 words, the final output might only be 800 to 900 words. 精确字数: 如果用户想用大语言模型写一个字数为1000的小说，可能最后生成的小说只有800-900字。
Content compliance: The content of user's AI large language model is subject to domestic regulations. When sensitive information exists Prompt question or AI answer, the model's manufacturer will apply regulations to unrevealed results. 内容合规: 用户使用的人工智能大语言模型的内容必须遵守国内的法规。如果出现敏感信息，无论是在用户的问题中还是在人工智能的回答里，模型的制造商都会按照规定处理，不对外公开这些信息。
Agents are tools for users to generate partnership scenarios which include structural guidance assistants, organisational assistance with agile applications. Many public agents include work, study, entertainment and creativeness aspects facilitating individuals to be an effective entity.
智能体是用户 生成伙伴关系场景的工具，包括结构化指导助手以及具有敏捷应用的组织协助。公开智能体涵盖工作、学习、娱乐和创新等多方面,促进个体成为一个有效的实体。
The benefits of agents can: 使用智能体的优点包括:
- use Prompt structured templates to adhere specific capabilities of AI models conclusively 使用 Prompt 结构化的模版将 AI 模型的特定能力固定，一劳永逸
- reduce input, repetitive operations and cognition 减少输入，反复操作和思考认知
- use adjusted parameters directly provided by AI assistant (Increase productivity) 使用已经调整好参数的 AI 助手所提供的服务（提高生产力）
- share among users to exercise and resolve easily 将助手分享给其他用户共同体验，轻松解决问题
Wen Xin Yiyan, Xunfei Spark and Kimi have set up many agents for users to input demands instead of structural prompts that can greatly reduce repetitive workload and improve efficiency.
文心一言, 讯飞星火, kimi都设置了很多智能体供用户使用,这些智能体使得用户可以简单输入诉求即可,不再需要结构化的提示词。这大大减少了重复工作量, 有效提升了工作效率。
Users can simply call these agents to complete tasks and create own agents from experience. Please visit details at (https://xinghuo.xfyun.cn/botcenter/createbot) 用户可以直接调用这些智能体帮助完成工作。当用户有经验时可自行创建智能体。创建智能体的相关细节可到访链接: (https://xinghuo.xfyun.cn/botcenter/createbot)
1. Compare the generated content of different tools for the same Prompt word. Discuss dialogue skills and Prompt word optimization methods. Choose the best generated content.
2. Design a brief research proposal to support research in your area of interest based on the best generated content.
3. Design an intelligent agent to complete the related tasks.
Tips:
Wenxin Yiyan (https://yiyan.baidu.com/)
Kimi (https://kimi.moonshot.cn/)
Xunfei Spark (https://xinghuo.xfyun.cn/desk)
1. 比较不同的工具针对同一个提示词的生成内容,讨论对话技巧和提示词优化方法。选择认为质量最高的生成内容。
2. 结合质量最高的生成内容, 设计一个简要的研究计划书, 用来支持感兴趣领域的研究。
3. 设计一个智能体,来辅助完成相关工作。
提示:
文心一言 (https://yiyan.baidu.com/)
kimi (https://kimi.moonshot.cn/)
讯飞星火 (https://xinghuo.xfyun.cn/desk)
AIGC tool does not limit to text dialogue. Wenxin Yiyan, Xunfei Spark and Kimi have uploaded attachments functions. When upload is completed, the AI large language model (AIGC tool) can generate parsed content.
生成式人工智能工具并不局限于文本对话。文心一言,讯飞星火, kimi都设置了上传附件的功能。上传附件后, AI 大语言模型(生成式人工智能工具)可以生成解析的内容。
For example: Kimi supports texts summarization, resolve language barriers to analyse literatures, reports, contracts, web pages by uploading PDF, Word, Excel, TXT files and images, lengthy documents and multiple files. The interpretation of Chinese novel is shown below.
例如: Kimi支持文本摘要，支持通过上传PDF、Word、Excel、TXT文件,图片以及长篇文档和多文件等来分析文献、报告、合同、网页。这打破了语言障碍。下图展示了用kimi解读中文小说的结果。
Now, try Kimi to "read" papers. 现在, 尝试用Kimi"阅读"论文吧。
For Kimi, please refer to (https://blog.csdn.net/daniellq/article/details/139724147)
关于kimi的具体使用,可以参考链接 (https://blog.csdn.net/daniellq/article/details/139724147)
Academic institution have specifications to regulate generative AI tools, such as ChatGPT usage since it has reduced the barrier of academic papers fraud moderately.
学术机构已经设立规范来监管生成式人工智能工具的使用，比如ChatGPT，因为它已经慢慢降低了学术论文造假的门槛。
Many Chinese academic publications have imposed restrictions on using generative AI tools to write journals, it is academic misconduct if is found.
许多中国学术出版商都限制使用生成式人工智能工具撰写期刊，一旦发现，将认定为学术不端行为。
Foreign academic publications have standardization. Journals such as Nature, Cell, The Lancet, and JAMA have stated that AI does not qualify as an author, and researchers using generative AI tools must declare in manuscripts.
国外学术出版物有相关规范. Nature（自然）、Cell（细胞）、The Lancet （柳叶刀）、JAMA（美国医学会杂志）等期刊均发表声明称，人工智能不具有作者资格，使用生成式人工智能工具的研究人员应在稿件中进行说明。
Please refer AIGC regulations by the Ministry of Science and Technology of China and Guideline on the Boundaries of AIGC Usage in Academic Publishing at: (https://www.most.gov.cn/kjbgz/202312/t20231221_189240.html)
(https://www.istic.ac.cn/html/1/245/1701698014446298352.html).
请阅读中国科技部的相关规定以及学术出版中AIGC使用边界指南链接: (https://www.most.gov.cn/kjbgz/202312/t20231221_189240.html)(https://www.istic.ac.cn/html/1/245/1701698014446298352.html)
1. Use Kimi to read, catalogue journals, and write literature review or notes with content generation. Kimi (https://kimi.moonshot.cn/)
2. Combine 2. Talk to AIGC related work to complete a full research paper. Discuss findings with peers.
3. What is the role of AIGC in academic research? Should it be banned for academic misconduct? Discuss AIGC usage in academic research.
1. 使用kimi辅助进行论文阅读以及文献整理等工作, 结合生成内容,尝试撰写文献综述或读书笔记。Kimi (https://kimi.moonshot.cn/)
2. 结合2.Talk to AIGC的相关工作, 完成一篇完整的研究论文, 及讨论发现。
3. 怎样看待AIGC在学术研究中的作用? 学术界是否应该全面封杀AIGC来避免学术不端？阐述对学术研究中AIGC使用的看法，讨论并整理在学术研究中AIGC使用的正确方式。
AI Prompt 工程师认证和学习指南: https://datawhaler.feishu.cn/wiki/BhVQw3FlFitUTAkS4oXcSyJanph OpenAI cookbook: https://cookbook.openai.com/
ChatGPT Prompt Engineering for Developers: https://www.deeplearning.ai/short-courses/chatgpt- prompt-engineering-for-developers/
ChatGPT Prompt Framework: https://learningprompt.wiki/docs/chatGPT/tutorial-extras/chatGPT- prompt-framework
Prompt Engineering Guide: https://www.promptingguide.ai/
文心一言使用手册: https://yiyan.baidu.com/learn
SparkDesk使用指南: https://www.xfyun.cn/doc/spark/Guide.html https://segmentfault.com/a/1190000044605356
© 拷贝以及使用本教程应得到讲师同意。请不要将本教程上传至互联网。