大语言模型在教育中的应用

各种 AI 教育工具： https://aieducator.tools/

搜索 AI 工具 https://theresanaiforthat.com/

MIT AI Edu https://raise.mit.edu/ https://raise.mit.edu/research-projects/

https://huggingface.co/MerlynMind/merlyn-education-corpus-qa-v2 https://huggingface.co/TheBloke/merlyn-education-corpus-qa-v2-AWQ 13b parameter decoder-style transformer model for the education domain. It is fine-tuned from a llama2-13b base-model.

精准学

AI辅学机Bong

精准学自主研发的“心流知镜”大模型中也结合了阿里云通义大模型。精准学自主研发的“心流知镜”大模型，结合阿里巴巴的千亿级通义模型。在研发过程中，通过16万小时的教育语音数据训练，能够自然地模仿超过20种个性化教学风格，具备多模态和个性化交互能力。

精准学创始人兼CEO杨仁斌曾任淘宝首席产品经理，也是淘宝“猜你喜欢”的首任产品经理。他认为，“在AI大模型技术诞生之前，从原理上讲，大家都做不出真正合格的学习机。因为有效学习必须同时解决三个问题，一是孩子坐得住，二是学习内容个性化，三是看视频就能学得会。后面两个相对好解决，但第一个几乎无解。家长现在购买学习机最担心的就是‘吃灰’。从2018年起，我们在教学机构场景中，在老师监督下使用精准学学习机，有着很好的学习效果。如今，随着AI技术突破，我们通过跟孩子及AI老师的全程连线和实时通话，终于同时解决了这三个问题，才有机会创造出合格的产品。”

Khanmigo

2024-04-22 POWER4EDU

Khanmigo的学术论文反馈功能，可以为学生的论文提供具体的和可操作的反馈。这包括了及时和有用的建议、论文结构介绍、适当的学术基调等。Sal Khan表示，在一年内，我们将有一位老师能够与人工智能一起分配工作。构建一个包含论文标题的写作活动，并通过人工智能来进行分配。人工智能不仅能提供反馈，还能在这个过程中提供帮助。比如说，“这是一篇论文，我们一起写提纲吧。”、“让我们找一些数据来支持你的论文观点。”然后，当人工智能向老师汇报时，它不会只是给出论文的评分。而是会给出“这篇文章花了四个小时。我们在写论文时遇到了一点麻烦，但我们最终还是做到了。”

借助GPT-4的能力，Khanmigo可以进行更加复杂的对话，为学生提供更加逼真的AI导师。Sal Khan表示，GPT-3.5的能力确实无法推动有意思的对话。如果一个学生说，‘嘿，告诉我答案’。如果是使用GPT-3.5，即使你告诉它请不要告诉我答案，它仍然会给出答案。

相反，Khanmigo会通过询问学生是如何得出答案的，或者指出他们是如何在数学问题上偏离方向的，来帮助学生自己找到答案。比如，Khanmigo能够给出“我们能够做到的是，‘做得好。看起来你把负2分配错了，为什么不再试一次呢？”或者，“你能帮我解释一下你的推理吗，因为我觉得你可能犯了一个错误？’”

可汗学院创始人Sal Khan认为，Khanmigo的幻觉现象仍有可能发生，但GPT-4模型已将错误概率降低很多

https://www.imconsultancy.net/power4edu/%E5%8F%AF%E6%B1%97%E5%AD%A6%E9%99%A2%E5%88%9B%E5%A7%8B%E4%BA%BASal%20Khan%E8%AE%A4%E4%B8%BA%EF%BC%8CKhanmigo%E7%9A%84%E5%B9%BB%E8%A7%89%E7%8E%B0%E8%B1%A1%E4%BB%8D%E6%9C%89%E5%8F%AF%E8%83%BD%E5%8F%91%E7%94%9F%EF%BC%8C%E4%BD%86GPT-4%E6%A8%A1%E5%9E%8B%E5%B7%B2%E5%B0%86%E9%94%99%E8%AF%AF%E6%A6%82%E7%8E%87%E9%99%8D%E4%BD%8E%E5%BE%88%E5%A4%9A

Google Generative AI for Educators

教老师 写 Prompt

https://grow.google/ai-for-educators/

##

https://ua-data7.github.io/introllms/education/#example-2-lecture-preparation

Teaching and learning with ChatGPT and Bard¶ ChatGPT and Bard can improve teaching and learning processes by generating and assessing information and can be used as a standalone tool or integrated into other systems. It can perform simple or technical tasks and examples show how it can augment teaching and learning.

Role playing Description Example of implementation Possibility engine AI can suggest alternative ways to express an idea Students can write queries in ChatGPT/Bard and use the “Regenerate” response function to explore alternative responses. Socratic opponent AI can act as an opponent to develop an argument Students can enter prompts into ChatGPT/Bard, using the structure of a conversation or debate. Teachers can ask their students to use ChatGPT/Bard to prepare for discussions. Collaboration coach AI helps groups to research and solve problems together When completing tasks and assignments, students can use ChatGPT/Bard to find information while working in groups. Guide on the side AI acts as a guide to navigating physical and conceptual spaces Teachers use ChatGPT/Bard to generate content for their classes or courses, such as discussion questions, and to seek advice on how to support students in learning specific concepts. Personal tutor AI tutors each student and gives immediate feedback on progress ChatGPT/Bard provides personalized feedback to students based on information provided by students or teachers (e.g., test scores). Co-designer AI assists throughout the design process Teachers can seek ideas from ChatGPT/Bard for designing or updating a curriculum, including rubrics for assessment. Alternatively, they can focus on specific goals, such as making the curriculum more accessible. ChatGPT can provide recommendations and suggestions to help achieve these objectives. Exploratorium AI provides tools to play with, explore, and interpret data Teachers provide basic information to students who write different queries in ChatGPT to find out more. ChatGPT/Bard can be used to support language learning. Study buddy AI helps the student reflect on learning material Students explain their current level of understanding to ChatGPT/Bard and ask for ways to help them study the material. ChatGPT/Bard could also be used to help students prepare for other tasks (e.g., job interviews). Motivator AI offers games and challenges to extend learning Teachers or students ask ChatGPT/Bard for ideas about how to extend students’ learning after providing a summary of the current level of knowledge (e.g., quizzes, exercises). Dynamic assessment AI provides educators with a profile of each student’s current knowledge Students engage in a tutorial-style dialogue with ChatGPT/Bard, and then request that ChatGPT/Bard create a summary of their current knowledge for sharing with their teacher or for assessment purposes.

Opportunities of using ChatGPT and Bard as a learning tool¶

Enhance lessons. In low- and middle-income countries, teachers face the challenge of making a dense curriculum engaging. To help with this, ChatGPT/Bard can translate learning objectives into lesson plans, offer ideas for class preparation, and aid in creating new assignments and assessments. However, ChatGPT/Bard does not assist with delivery, so teachers must still have strong teaching skills to ensure quality lessons. Create assessment questions. ChatGPT/Bard can help teachers improve assessment questions and generate multiple-choice items. It can also encourage higher-order thinking skills by providing prompts for essay questions and practical tasks. By using different types of assessment, teachers can help students develop critical thinking, problem-solving, and collaborative skills. Support with language barriers. The ideal scenario is for teachers to instruct in their native language, but some education systems are shifting towards teaching in a second language, such as English, even when evidence suggests the opposite approach. In such situations, teachers who are not proficient in English struggle to teach effectively. A proposed solution is the use of chatbots to enhance teachers’ language proficiency, enabling them to teach better in both their native and a foreign language. Tools like Duolingo and ChatGPT reportedly provide affordable, accessible, and highly personalized language lessons. Provide additional support to students. ChatGPT/Bard can be used by teachers to encourage student curiosity and generate ideas for homework assignments. AI tools are particularly helpful in identifying the source information used in the chats. However, there is a risk that students may ask the chat to complete their homework for them instead of seeking help. To address this risk and teach about integrity, teachers can discuss the limitations of these tools, such as privacy risks, bias, and hallucination. Furthermore, teachers should focus on questions that cannot be answered by ChatGPT, such as those that require knowledge outside of the chatbot’s training data, such as human emotions or subjective perspectives. Grading assessment and papers. ChatGPT/Bard can be used to automatically grade multiple-choice/one-answer tests; it can also help teachers with standards-based grading. This has potential unintended consequences for low accuracy, poor grading, or false positive proctoring (mistakenly red-flagging students for cheating). When considering systems for proctoring or grading, it is critical to take measures to secure fairness, accountability, confidentiality, and transparency of their algorithms whenever needed.

Practical AI for Instructors and Students

沃顿商学院

Part 1: Introduction to AI for Teachers and Students Part 2: Large Language Models (LLMs) Part 3: Prompting AI Part 4: AI for Teachers Part 5: AI for Students https://www.youtube.com/playlist?list=PL0EdWFC9ZZrUAirFa2amE4Hg05KqCWhoq

## Creative ideas to use AI in education 很多老师的想法。很宝贵。 https://docs.google.com/presentation/d/1wVgLWgeEvJm3fznlm0aV8ZiuWsW3o3aUQUCcvuM5vxQ/edit#slide=id.g229cc0973b2_3_0

Vanderbilt University

https://www.coursera.org/learn/chatgpt-innovative-teaching

Guidelines for Syllabus Statements About Generative AI，August 2023 v 2.1.1，课程大纲中，如何提出对 生成式 AI 的 要求：完全禁止，部分，还是允许，https://artificialintelligence.arizona.edu/sites/default/files/2023-08/syllabus%20guidelines%20for%20AI%20v2.1.1-copyable.pdf

https://ua-data7.github.io/introllms/education/#chatgpt-and-bard-used-in-academic-and-essay-writing

8 个视频 • 总共 65 分钟 通过 ChatGPT 了解教育创新概述•16 分钟•预览单元 快速备课和集思广益•12 分钟 建立个性化和有吸引力的范例•8 分钟 创建吸引学生的活动•6 分钟 思考解决问题的其他方法•7 分钟 创建教育游戏 I•5 分钟 学员按需练习•4 分钟 预评估，学生应用评分标准进行反馈•4 分钟 8 个阅读材料 • 总共 45 分钟 导言•10 分钟 什么是生成式人工智能？•10 分钟 示例生成的提示模式示例•3 分钟 活动生成的提示模式示例•3 分钟 生成解决方案的提示模式示例•3 分钟 按需练习的提示模式示例•3 分钟 创建预评估的提示模式示例•3 分钟 保持联系，了解更多•10 分钟 1 个作业 • 总共 20 分钟 制作有趣的教育材料•20 分钟

Demo

谷歌的 Gemini 模型可以读入手写的数学和物理习题，检查其中的公式哪些对，哪些错，错在哪里，并且给出一步一步的推导过程；可以让它提供进一步的解释，它就可以提供关于一个话题的个性化解释；还可以要它提供进一步的练习题，它就提供个性化的练习题 Youtube 视频。

Exploring artificial intelligence with StoryQ, Website, why Artificial Intelligence Belongs in English Class.

教育 AI 系统设计

要设计一个好的教育 AI 程序，需要了解一些基本的教育理论和方法，评估一些现有的 AI 教育应用，找到自己项目要解决的问题，并进行设计。

为此，我们下面学习 MIT 6.S062/MAS.S10/MAS.s60 Generative Artificial Intelligence in K12 Education 课程。该课程会带我们走过上述过程，最后利用 ChatGPT、DALL-E 等生成式 AI 工具，开发一个面向 K12 学生的教育软件，并发表一篇论文。

该课程的具体内容如下：

第一课，熟悉 OpenAI API。用 OpenAI API 创建一个“猜人名”的游戏。注意，它代码里的 API 比较老了。在 Colab 里运行时，会被提示安装较低版本的 OpenAI 库。虽然如此，但这个代码还是有价值，适合入门。

第二课，介绍 ChatGPT 原理和 4 个学习理论：合作、互教、组内成员互补/多样化、高质量/适合学生需求的无处不在的反馈。然后，头脑风暴，提出项目创意。

第三课，理解生成式 AI 的 Bias、意想不到的结果，学习如何减少 Bias；学习如何分析系统的相关各方，建立 Ethical 矩阵；学习 Prompt 工程，生成图像的正/反方法。

第四课，学习联合国教科文组织 UNESCO 的面向学生 AI 能力的框架。这个框架包括 5 个方面：Mindset、Ethics、基础、技术、解决问题。每个方面包括 3 个层次：理解、应用、创造。明确 AI 教育的目的。

第五课，进行用户可用性测试，即先头脑风暴，用户的需求，然后召集志愿者，调研；最后进行测试，Think Aloud。

第六课，学习如何培训教师。了解成年人学习的特点。成年人在学习前，需要知道为什么要学；孩子会更探索、玩耍一些，也更能从结构化的设计中受益。

第七课，评估四套 AI 课程，和学校的一线教师一起讨论，用起来是否方便、内容、是否适合课堂教学。

最后是项目，我们可以看到麻省理工的同学们最后作品的视频、PPT 和论文。挺有意思的。这就是一个完整的基于 ChatGPT 的项目。

请点击上面的网页上的链接，获得各部分的材料，进行学习。

项目

我们下面基于学习到的教育理论和示例软件，开发自己的系统。下面是一个可能的系统，看大家是否感兴趣。

这个系统能够培养珍贵的普通人；具体来说，为初入职场和社会的年轻人，逐渐走向自立，提供支持，主动、求真、双赢。

实现上来说，包括两部分：

第一部分是知识和技能，具体来说，知识包括：认识自我、职场、职业、世界；技能包括：目标管理、时间管理、计划执行、专业能力、职场关系、双赢思维、真诚沟通

第二部分是不断练习这些知识和技能，通过 5 步，不断解决现实生活中的问题：理解问题和挑战、确定目标、制定计划、落实计划、完成和复盘，最终实现自立。

企业开发经验分享

我们现在开发了一个小系统。那么，在 Github 这样的大公司，他们是怎么开发系统的呢？和我们上面做的其实差不多。下面是 GitHub Copilot 团队分享的他们构建 LLM 应用程序的经验。包括三步：首先定义问题，然后迭代，创造平滑的 AI 产品经验，最后优化质量、安全。我们 Build LLM 的应用，也要走这个流程：How to build an enterprise LLM application: Lessons from GitHub Copilot，

会议

参考

系统

研究

https://www.moreusefulthings.com/resources

Dr. Ethan Mollick is an Associate Professor at the Wharton School of the University of Pennsylvania, where he studies and teaches innovation and entrepreneurship, and also examines the effects of artificial intelligence on work and education. His academic papers have been published in top management journals and his research has been covered by CNN, The New York Times, and other leading publications. His newest book on AI, Co-Intelligence, is out on April 2, 2024.

Current papers include:

New Modes of Learning Enabled by AI Chatbots: Three Methods and Assignments.

In this paper, we discuss the opportunity provided by AI because it can help us teach in new ways. The AI’s flaws —its tendency to make up facts, its lack of nuance, and its ability to make excellent student essays — can be used to improve education. This isn’t for some future theoretical version of AI. Instructors can create assignments right now, using ChatGPT, that will help stretch students in new ways.

Using AI to Implement Effective Teaching Strategies in Classrooms: Five Strategies, Including Prompts

In the rush to deliver AI benefits directly to students, the role of teachers is often overlooked. AI tutors, as exciting as they are, do not replace the complex role of a teacher in front of a class. But not enough effort seems to be going toward applying AI to help instructors. We have a new paper that tries to remedy that gap by providing some research-backed approaches to pedagogy and the AI prompts (for GPT-4, GPT-3.5, and other AIs) to implement them

Assigning AI: Seven Approaches for Students, with Prompts

The incredible promise of AI as a way for students all over the world, of all ability levels, to learn is undeniable. Education is our most powerful system for increasing social mobility, unlocking potential, and improving lives. A tool that can help with this has tremendous implications. Plus, students are already using AI for direct help. Teaching them how to do it responsibly may alleviate some of the negative implications of our AI moment. In this paper, we tackle ways that students can be assigned to use AI directly. We don’t shy away from the dangers but provide detailed instructions on how students and instructors can think about each of the tools we suggest.

Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality (Dell’Acqua, Fabrizio and McFowland, Edward and Mollick, Ethan R. and Lifshitz-Assaf, Hila and Kellogg, Katherine and Rajendran, Saran and Krayer, Lisa and Candelon, François and Lakhani, Karim R)

In our study conducted with Boston Consulting Group, a global management consulting firm, we examine the performance implications of AI on realistic, complex, and knowledge-intensive tasks. The pre-registered experiment involved 758 consultants comprising about 7% of the individual contributor-level consultants at the company. After establishing a performance baseline on a similar task, subjects were randomly assigned to one of three conditions: no AI access, GPT-4 AI access, or GPT-4 AI access with a prompt engineering overview. We suggest that the capabilities of AI create a “jagged technological frontier” where some tasks are easily done by AI, while others, though seemingly similar in difficulty level, are outside the current capability of AI. For each one of a set of 18 realistic consulting tasks within the frontier of AI capabilities, consultants using AI were significantly more productive (they completed 12.2% more tasks on average, and completed task 25.1% more quickly), and produced significantly higher quality results (more than 40% higher quality compared to a control group). Consultants across the skills distribution benefited significantly from having AI augmentation, with those below the average performance threshold increasing by 43% and those above increasing by 17% compared to their own scores. For a task selected to be outside the frontier, however, consultants using AI were 19 percentage points less likely to produce correct solutions compared to those without AI


Index Previous Next