什么是生成式 AI？

生成式 AI 是一个术语，适用于使用数学模型创建新事物的工具。示例包括 Chat GPT、Gemini、DALL-

E、Midjourney、Perplexity 和 Claude。这些工具利用了人工智能研究和大量硬件的最新进展。与以前只能找到已
经存在的东西的检索 AI 工具（通常在搜索引擎中找到）不同，生成式 AI 可以“创建”独一无二（尽管公式化）的项目——诗
歌、图像、歌曲、散文。虽然不可否认令人印象深刻，但结果可能并不总是 100% 准确，新内容可能会重复任何偏差，并将反映
训练数据中现有的知识差距。

Generative AI is a term applied to tools that create new things using mathematical
models. Examples include Chat GPT, Gemini, DALL-E, Midjourney, Perplexity and
Claude. These tools take advantages of recent advances in artificial intelligence
research and masses of hardware. Unlike previous retrieval AI tools (often found in
search engines) that could only find things that already existed, generative AI can
"create" items - poems, images, songs, essays - that are unique (albeit formulaic).
Whilst undeniably impressive, the results may not always be 100% accurate, and the
new content may repeat any bias and will reflect existing knowledge-gaps in the
training data.

Can I use it in exams and assessed work (is it cheating?)

At all times you should be submitting material that is in your own words and the
result of your own thinking. Submitting unedited AI-generated answers is
cheating and may incur academic misconduct procedures. The University has
decided that generative AI can be used in certain assessments. Staff should make it
clear to you whether its use is allowed in any given assessment and if so, whether
you need to record and acknowledge if and how you use it.

在任何时候，您都应该提交用您自己的话和您自己思考的结果编写的材料。提交未经编辑的 AI 生成的答案是作弊，可能会导致学
术不端行为程序。大学已决定生成式 AI 可用于某些评估。工作人员应向您明确说明在任何给定的评估中是否允许使用它，如果允
许，您是否需要记录并确认您是否以及如何使用它。

有大学认可的工具吗？
是的，您可以使用 Microsoft Copilot - 请先登录以获得额外保护。我们还提供了一些建议问题，供您在自己尝试第三方工
具之前询问，以确保您和您的数据安全。

What is AI？

Artificial intelligence (AI) is a term applied to computer systems that are able to
perform tasks commonly associated with intelligent beings. Tasks may require
complex logic - such as playing chess - or pattern recognition - e.g. identifying people
in photographs, or writing stories.

What is "generative AI"?

Generative AI is a term applied to tools that create new things using mathematical
models which draw from artificial intelligence research. Machine learning
algorithms, capable of manipulating billions of parameters simultaneously, trained
by processing vast amounts of data, can produce "new" content using relationships it
has identified in the content that it has seen before. Whilst undeniably impressive,
the results may not always be 100% accurate, and the new content may repeat any
bias or knowledge-gaps in the training data.

That may still seem a bit abstract, so we'll unpick this more in this section and again
in How does it work?

如何在大学中使用 AI？
Ways you could use generative AI

Listed below are some suggested ways that you might want to use generative AI
tools in your studies. Note at all times you need to be careful and ask yourself who is
doing the work and, most importantly, who is doing the thinking, making the
decisions and the connections?

Overcoming writer's block

We all experience moments where we don't know where to start on a question and
stare at a blank screen or page. Generative AI can be a useful tool for getting over
this, just like a normal search engine. If there are key words in the assignment title -
such as "evaluate" - use a tool like ChatGPT to help clarify their meaning. Use it to
generate different perspectives on an issue, suggest possible structures or ideas to
pursue. Do this with care - remember it generates answers based on probability, not
from any underlying consciousness. Use the content to inform the way you structure
your own response and ensure that the final product is based on your own work
drawn from multiple reliable sources.

If you are working on a long project, you could also try and use it to help with your
time management - create todo lists and plans, or break the task down into smaller
bits.

Providing explanations and summaries

Large language models are very good at rewording and summarising. As such, if
there is something you are struggling with - perhaps a section of postmodern
critique - then use a generative AI tool to provide you with a summary or simplify the
text for a non-specialist reader. Treat this as a starting point - it does not always get
things right - and use this new insight to check facts and definitions in a reliable
textbook or in discussions with fellow students or your tutor.

There are some tools that offer to provide a summary of papers. We don't
recommend using these as a short-cut to reading and understanding the article! You
risk breaching copyright regulations (e.g. by uploading a PDF version of an online
journal to a third party site). You will also miss out on the detail and methods and
there is a real danger that you'll misunderstand the paper's conclusions if the
generative AI oversimplifies things or just gets these wrong! There is no substitute
for reading and thinking about the original work.

Polish your critical thinking skills

The best way to develop your critical thinking is to discuss ideas with other people.
Some people find that a daunting task, so why not practice with AI? Use generative
AI tools to provide different responses to the same question. Then apply your own
critical thinking skills to evaluate the output. Check the accuracy of the information
provided using other sources and see if you can associate the arguments with any
particular thinkers. Search for materials that can support or counter the arguments
in each prompt. You could even enter further prompts containing your own counter-
arguments and ask the tool for feedback on their validity and strength of argument!

Drafting and experimenting with different structures

If you have taken copious notes from relevant materials and are now trying to
combine these into a unified narrative, then tools such as Copilot, Gemini and
ChatGPT can help. Enter the materials into the prompt and ask it to generate a
summary, drawing upon your findings. If your response has a required structure (e.g.
introduction, methods, data, conclusion) then include this information in the prompt
too. This may give you some clues towards a possible structure for your final essay.

Note we strongly recommend that you ask it to generate several potential

structures. You should then read through each structure to see if it really does apply
to the topic and adequately reflect the topic's complexity.

Improving your spelling, grammar and structure

Large language models have been trained on a huge corpus of text, which makes
them very good at spelling and grammar. Note though that the output can be biased
towards English and often US-English. As such, always pass the results through spell-
checking and grammar tools afterwards, when you incorporate the results in your
draft. Don't copy and paste the results straight into your document, instead look at
the changes it has suggested and use them to re-write your own text.

Using different writing styles

Sometimes assignments will require you to write for a particular audience - perhaps
in the style of a journal article where you can assume a high level of technical
knowledge, or a press briefing, aimed at the general public. If you are struggling to
choose an appropriate style, try putting some of your writing through a generative AI
tool and ask it to rewrite it for the specified audience. That should give you some
tips. Or you may find it difficult to condense your writing to meet the word limit.
Generative AI tools can help you identify the key issues that you need to keep. Be
careful when you come to write your final piece that the meaning and conclusions
are still valid and the words you use are yours.

Create media

You could use generative AI tools to create images, graphics, and other visual
content to use in a presentation, or as part of a report. This can be particularly handy
if you want to create images that are hard to find online, perhaps challenging
stereotypes that all scientists must wear white coats. Don't forget to cite the source
(and add alt text)!

Coding

There are lots of code-samples on the web and so large language models are quite
good at writing simple code (particularly high level languages like Python). If given a
task, most tools will offer some suggested code alongside an explanation of what
each step does.

Treat these samples just like others you'd come across on websites or textbooks -
often they work, but occasionally they may make mistakes, and they may use
obsolete libraries or do things in a very inefficient manner. Always test them out,
take time to understand how they work and ask yourself whether there are ways
that the code could be improved to make it run faster, be less error-prone or easier
to understand.

Use it to create study resources

Generative AI tools are good at summarising and identifying patterns. As such you
might want to use them to help collate your notes, or turn them into flash-cards to
help you revise. You could even ask it to generate exam-style questions based on
your notes to help you practice responding to unseen exam questions or for oral
exams.

Use it as a writing coach

Generative AI is not a subject expert like your lecturers, but you can still ask it for
feedback on a draft. This may highlight areas where your writing is unclear or poorly
structured. If possible, provide details of the assessment criteria (e.g. a rubric) in the
prompt as this will help structure the responses it gives.

Note that this feedback will be generated by a mathematical model not a person. As
such it won't be aware of the context of the assessment, recent developments in the
discipline, what you have been taught and what you are expected to know at this
stage. Even if you ask it for a mark, that will not be the same as one generated by a
person. Use the feedback as a guide, but not as the final judge - you have to decide
when your work is ready to submit and how confident you are it answers the
question.

克服写作障碍
我们都有过不知道如何开始回答问题，只是盯着空白屏幕或纸张的时刻。生成式 AI 可以成为帮助克服这种情况的有用工具，就像
普通的搜索引擎一样。如果作业标题中有关键词——比如“评估”——可以使用像 ChatGPT 这样的工具来帮助澄清它们
的含义。使用它来生成对某个问题的不同视角，建议可能的结构或需要探讨的想法。这样做时要谨慎——记住，AI 生成的答案
是基于概率的，而不是出自任何真正的意识。使用这些内容来帮助你构建自己的回应，确保最终的成果基于你从多个可靠来源获得
的独立工作。
如果你在进行一个长期项目，也可以使用 AI 帮助你进行时间管理——创建待办事项清单和计划，或者将任务分解为更小的部分。

提供解释和总结
大型语言模型在改写和总结方面非常擅长。因此，如果你在某些方面感到困难——例如理解后现代批评的某一部分——可以使
用生成式 AI 工具为你提供总结，或将文本简化为适合非专业读者的内容。将其作为起点——它并不总是完全正确的——利
用这种新获得的见解在可靠的教科书中检查事实和定义，或与同学或导师讨论。
有些工具提供文献的摘要功能。我们不建议将这些作为阅读和理解文章的捷径！这样做可能违反版权法规（例如，上传 PDF 版本
的在线期刊到第三方网站）。你还可能错过细节和方法，如果 AI 过于简化或出错，甚至可能误解文章的结论！没有什么能代替阅
读和思考原始作品。

打磨你的批判性思维能力
发展批判性思维的最好方法是与他人讨论想法。有些人可能觉得这是一项艰巨的任务，那为什么不和 AI 练习呢？使用生成式 AI
工具为同一个问题提供不同的回答。然后运用你的批判性思维能力来评估这些输出。使用其他来源来验证所提供信息的准确性，看
看你是否能将这些论点与某些特定的思想家联系起来。查找能支持或反驳每个 AI 输出的材料。你甚至可以输入自己的反驳论点，
询问 AI 它们的有效性和论证的力度！

起草和实验不同的结构
如果你已经从相关资料中做了大量笔记，正在尝试将这些资料组合成一个统一的叙述，那么像 Copilot、Gemini 和
ChatGPT 这样的工具可以提供帮助。将这些材料输入提示，要求它基于你的发现生成一个总结。如果你的文章需要特定的结构
（例如，介绍、方法、数据、结论），那么也应在提示中包含这些信息。这可能会为你的最终文章提供一些结构上的线索。
请注意，我们强烈建议你让 AI 生成多个可能的结构。然后你可以逐一阅读每个结构，看看它是否真的适用于该主题，并充分反映
出该主题的复杂性。

改进拼写、语法和结构
大型语言模型经过了大量文本的训练，因此它们在拼写和语法方面非常擅长。然而，生成的内容往往偏向英语，尤其是美式英语。
因此，当你将其结果融入草稿时，最好再通过拼写和语法检查工具进行复核。不要直接将 AI 的结果复制粘贴到文档中，而是观察
其建议的更改，并使用它们重新撰写自己的文本。

使用不同的写作风格
有时作业会要求你针对特定的受众写作——例如，以期刊文章的风格写作，假设读者具有较高的专业知识，或写成新闻发布会简
报，面向公众。如果你在选择适当的写作风格上遇到困难，可以将你的一部分写作输入生成式 AI 工具，要求它为特定受众重写。
这可能会给你一些提示。或者，当你需要精简写作以符合字数限制时，生成式 AI 工具可以帮助你确定需要保留的关键问题。在撰
写最终版本时，一定要注意确保意义和结论仍然有效，且所使用的文字是你自己的。

创建媒体
你可以使用生成式 AI 工具创建图像、图形和其他视觉内容，用于演示或报告的一部分。如果你想要创建在线难以找到的图像，这
尤其有用，比如打破所有科学家都必须穿白大褂的刻板印象。不要忘记引用来源（并添加替代文本）！

编写代码
网络上有很多代码示例，因此大型语言模型在编写简单代码方面表现不错（特别是像 Python 这样高级的语言）。在给定任务时，
大多数工具会提供一些代码建议，并解释每一步的作用。
将这些代码示例与在网站或教科书上看到的代码一样对待——它们通常有效，但有时会出错，可能使用过时的库，或者以低效的
方式完成任务。始终测试代码，花时间理解其工作原理，并考虑是否有改进的方法，以使代码运行得更快、出错更少或更易于理解。

用 AI 创建学习资源
生成式 AI 工具擅长总结和识别模式。因此，你可以利用它们来帮助整理笔记，或将其转化为帮助你复习的闪卡。你甚至可以让它
根据你的笔记生成考试风格的问题，帮助你练习回答未见过的考题或准备口头考试。

用作写作教练
生成式 AI 不是像你的讲师那样的学科专家，但你可以向它请求对草稿的反馈。这可能会指出写作中不清楚或结构不佳的地方。如
果可能的话，在提示中提供评估标准的细节（例如评分标准），这将有助于 AI 构建它的回应。
注意，这些反馈是由数学模型生成的，而不是由人生成的。因此，它不会意识到评估的背景、学科的最新发展、你所学习的内容以
及你在此阶段应该知道的内容。即使你让它给出一个分数，那也不会与真人评分相同。将反馈作为指南，而非最终裁判——你必
须决定你的工作何时准备好提交，以及你对它是否回答了问题感到有信心。

Using AI generated content

Be critical
Take good notes

是否有大学批准的生成式 AI 工具？
Yes
Staff and students at Durham University all have access to Microsoft Copilot（在新选项卡中打开） via the
Edge browser. (Note that this is not the same as the confusingly named Microsoft 365 Copilot（在
新选项卡中打开） which would connect to all the data you can see in Outlook, OneDrive, SharePoint,
etc.). Microsoft Copilot doesn't do that, but it provides access to content via Microsoft's Azure
OpenAI.
Accessing it
You can access Microsoft Copilot in the Edge browser from computers managed by the
University or on your own device. You can also access Copilot from any browser
via www.microsoft.com/copilot（在新选项卡中打开） (you may be asked to log in before it appears).
This tool combines generative AI tools with Microsoft's Bing web search tool (a process
sometimes referred to as "web grounding"), so it provides content complete with sources. They
still need checked and should be approached critically!

是的，达勒姆大学的教职工和学生都可以通过 Edge 浏览器访问 Microsoft Copilot。（注意，这与名称容易混淆的

Microsoft 365 Copilot 不同，后者可以连接到你在 Outlook、OneDrive、SharePoint 等中看到的所有数
据）。Microsoft Copilot 并不会这样做，它是通过微软的 Azure OpenAI 提供内容访问的。

访问方式
你可以通过大学管理的电脑或自己的设备在 Edge 浏览器中访问 Microsoft Copilot。你也可以通过任意浏览器访问*
*www.microsoft.com/copilot**（可能需要登录后才能显示）。

该工具将生成式 AI 工具与微软的 Bing 网络搜索工具结合使用（这种过程有时被称为“网络支撑”），因此它提供的内容附带来

源。不过，仍需仔细检查这些来源，并以批判的态度对待！