Chapter 4. Deep Research and AI-Assisted Academic Writing

1. Collecting and Analyzing Academic Materials with the Deep Research Feature

When doing homework or writing a report at school, the first thing you do is look for materials. If you enter keywords into an internet search box, thousands or tens of thousands of results pour out. Finding exactly what you need among such a vast amount of information is as difficult as finding a needle in a haystack.

Some articles are too old to be useful, and others are hard to verify as accurate. The "Deep Research" feature solves these problems. Deep Research means "in-depth investigation."

If a general search is like skimming the surface of the water, Deep Research is like diving deep underwater to find treasures on the bottom. AI platforms like ChatGPT, Perplexity, and Gemini provide this feature. Once you ask a question, the AI searches multiple times on its own to answer it, digging deeper based on the information it finds.

Let's look at an example of how Deep Research works. Suppose you are writing a report on the topic "The Impact of Climate Change on Korean Agriculture." A general search will give you a mix of news articles and blog posts.

However, with Deep Research, the AI first analyzes the topic. It divides it into sub-items such as "The Impact of Rising Temperatures on Rice Yield," "The Relationship Between Drought and Fruit Cultivation," and "Future Agricultural Response Strategies." Then, it reads dozens to hundreds of websites, papers, and reports for each item.

Based on what it finds in the first search, it creates a chain of follow-up questions to investigate even deeper. The advantage of Deep Research when collecting academic materials is that it prioritizes reliable sources.

Instead of unverified stories floating around on blogs, it first brings in papers written by university professors or researchers and statistical data released by government agencies. AI has the ability to judge the reliability of materials, so it distinguishes between "Primary Sources" and "Secondary Sources."

Primary sources are papers containing experimental results or original data. Secondary sources are review papers or textbooks that summarize the research of others. To write a good report, you should refer to many primary sources.

Here is how to write effective prompts when using Deep Research in the field. A prompt is a command given to an AI. If you vaguely say "Tell me about climate change," the results will also be vague.

You need to be specific to get a good answer. Deep Research Prompt Example 1 - Basic Type. Topic: The impact of artificial intelligence on the medical field.

Scope: English papers and Korean government reports published from 2020 to 2025. Question: How effective are AI diagnostic systems in increasing the early detection rate of cancer? Requirements: 5 key claims, 2 or more pieces of evidence for each claim, include source links and publication dates.

Exclusion: Do not use blog posts or promotional materials as evidence. Deep Research Prompt Example 2 - Comparative Analysis Type. Claim A: "Remote classes lower students' concentration."

Claim B: "Remote classes increase self-directed learning ability." Please find 3 research papers supporting each of these claims. Organize the research method, target population, and key conclusions of each paper into a table.

Sources are limited to materials published by university research institutes or academic journals. After collecting data, you move on to the analysis stage. Deep Research doesn't stop at just gathering materials; it compares multiple materials and finds commonalities and differences.

If you request, "Organize into a table the differences between the two claims, where Paper A says this and Paper B says that," it makes complex content easy to understand at a glance. Fact-checking is also an important role of Deep Research. There is a lot of false information on the internet.

If you ask the AI, "Check if this claim is true by comparing it with other reliable materials," it performs cross-validation. It doesn't trust information from just one place but confirms the same content across multiple sources. There are also things to be careful about when using Deep Research.

Not all information found by AI is correct. Sometimes AI speaks as if a non-existent paper actually exists; this is called "Hallucination." Therefore, you must personally search for the titles of papers or the authors' names provided by the AI to confirm they actually exist.

You should get into the habit of clicking the source links to verify the original text for yourself. In the end, Deep Research is a powerful tool that saves time. AI can process in a few minutes what would take a person a week to dig through in a library.

We can save the energy used for finding materials and focus instead on developing new ideas based on that data.

2. Organizing Research Using NotebookLM

No matter how much material you collect, it is useless if it is not organized. If papers are piled up like a mountain on your desk, it is hard to find what you need. Think of NotebookLM, an AI-based research tool created by Google, as your own smart research notebook and private tutor.

There is a key difference between NotebookLM and other AIs. General AIs like ChatGPT answer based on information from across the entire internet. As a result, they sometimes say strange things or make up facts that don't exist.

However, NotebookLM only reads and analyzes materials directly uploaded by the user. You can upload PDF files of papers you've found, notes from class, links to internet articles, and even YouTube videos. The AI studies only the content within those materials and finds evidence only from those materials when answering questions.

This is called "Grounding." Just as it sounds like having your feet firmly planted on the ground, it means answering based on solid evidence. Using NotebookLM is simple.

Access notebooklm. google. com in your web browser.

If you log in with your Google account, you can use it immediately without separate registration. Click the 'New Notebook' button and choose a name. Give it an easy-to-recognize name like "Final Exam Preparation" or "Thesis Reference Materials."

Then click the 'Add Source' button to upload the files you want to analyze. The free version allows you to upload up to 50 files per notebook. Including authoritative materials increases the accuracy of the results.

Authority here refers to the quality of being trustworthy. Statistical data released by government agencies, papers published in famous academic journals, and educational materials created by universities have high authority. On the other hand, blog posts by unknown authors or promotional articles have low authority.

The more authoritative materials you put into NotebookLM, the more accurate the summaries and analyses the AI produces. The "Garbage In, Garbage Out" principle applies here as well. What's even more important is when you put in materials you researched or experimented with yourself.

For example, suppose you experimented for a month on the topic "Will plants grow better if they listen to music?" You have a table summarizing the height of the plants, the number of leaves, and the amount of water given each day.

If you upload this table to NotebookLM and say "Find meaningful patterns in my experimental data," the AI will analyze your own unique data. Since it is material that only you have and is not found anywhere on the internet, it is information the AI is seeing for the first time. The analysis results from such materials become original insights.

NotebookLM Prompt Example 1 - Analyzing Authoritative Materials. Please find the core claims commonly mentioned in the 5 uploaded papers. For each claim, indicate the source number from which paper it came.

If there are parts where opinions differ, please organize those as well. NotebookLM Prompt Example 2 - Analyzing My Own Experimental Data. Please analyze the experimental data file I uploaded.

Compare the growth difference between the experimental group (plants that listened to music) and the control group (plants that did not listen to music). Explain if there is a statistically significant difference. If there is data that can be visualized as a graph, please recommend what kind of graph would be best.

NotebookLM Prompt Example 3 - Literature Comparison Analysis. Please compare the research methods of the uploaded Paper A and Paper B. Please create a table with the following items: - Number of research subjects - Experimental period - Measurement method - Main conclusions - Limitations.

Citation numbers are attached to NotebookLM's answers. For example, if it answers "Plants can respond to music [1]," clicking [1] will take you directly to the relevant part of the original document where that content appeared. This is the strength of NotebookLM.

You can see with your own eyes if that content truly exists in the document, rather than the AI just making it up. The Audio Overview feature is also interesting. NotebookLM creates an audio file based on the uploaded material, like a radio broadcast where two people are talking.

When you don't want to read dense text, you can listen to this audio to understand the content like a podcast. It's great for reviewing with earphones on your way to and from school. The Mind Map feature shows the relationship between various materials as a diagram.

The core topic is in the center, with detailed contents branching out to show connections. It helps you grasp the overall structure at a glance when your thoughts feel complex. Using NotebookLM doesn't mean reading completely disappears.

Rather, the order of reading changes. First, the AI draws a big map, and then we dive deep only where necessary. It's a method of looking at each tree after seeing the whole forest.

3. Collecting Materials via AI-Generated Search URLs and Gaining Insights from NotebookLM

A. Collecting Official Reports and Research Paper PDFs Using Google Search URLs. There is a vast amount of information on the internet, but what we really need is reliable material.

These include official reports released by government agencies, papers written by university research institutes, and statistical data created by international organizations. Most of these materials are in PDF format and can be found via Google. The problem is that many people don't know how to enter search terms to get the specific materials they want.

Google has an advanced search feature. If you enter special commands into the search box, you can filter for only the types of files you need. For example, if you type filetype:pdf, only PDF files will be searched.

If you type site:go. kr, it searches only on Korean government agency websites. Combining these commands allows for very accurate searching.

However, these commands are difficult to memorize. This is where AI helps. You can ask an AI like ChatGPT or Claude, "Make me a Google search URL to find the materials I want."

Since AI knows Google's advanced search syntax well, it can immediately create a search address that meets your specific criteria. Google Search URL Creation Prompt Example 1 - Finding Korean Government Reports.

Please create a Google search URL that meets the following conditions: - Topic: AI Ethics Policy - File format: PDF - Source: Korean government agencies (go. kr domain) - Period: Materials published after 2023 - Please provide it as a clickable link. The URL created by the AI looks like this: https://www.

google. com/search? q=AI+Ethics+Policy+filetype:pdf+site:go.

kr+after:2023. If you paste this address into your browser, only PDF files related to AI ethics policies announced by Korean government agencies after 2023 will appear.

You can easily find reports from authoritative organizations such as the Ministry of Science and ICT, the Personal Information Protection Commission, and the National Assembly Research Service. Google Search URL Creation Prompt Example 2 - Finding Overseas Academic Papers.

Please create a Google search URL that meets the following conditions: - Topic: climate change agriculture impact - File format: PDF - Source: Academic institutions (. edu domain or journal sites) - Language: English - Period: From 2022 to 2025. Google Search URL Creation Prompt Example 3 - Finding International Organization Statistical Data.

Please create a Google search URL that meets the following conditions: - Topic: AI adoption statistics - File format: PDF - Source: International organizations such as OECD, World Bank, IMF, UN - Period: Within the last 2 years. Download the PDF files from the search results and upload them to NotebookLM.

If you put in reports from several organizations at once, the AI will synthesize and analyze the entire content. Example of Official Report Analysis Prompt in NotebookLM: Please organize the following from the 5 uploaded government reports:

1. A table comparing the AI ethics principles proposed by each report

2. Three core values commonly emphasized across the reports

3. Areas where each report takes a different approach and the reasons why

4. Items mentioned as unresolved challenges

Please indicate the source number for each piece of information. The advantage of this method is that you are directly analyzing primary sources—the original reports and papers—rather than secondary materials like blogs or news articles.

Primary sources are the most accurate and reliable because they are the originals written directly by researchers or organizations, not something summarized or interpreted by someone else. B. Collecting International Seminar Presentations Using YouTube Search URLs.

Numerous academic seminars and conferences are held every day around the world. Valuable presentations, such as AI ethics lectures at Harvard University, technology innovation presentations at MIT, policy debates in the European Union, and robotics seminars in Japan, are uploaded to YouTube.

The problem is that these videos are in various languages such as English, Japanese, German, and French, and we may not know which search terms to use to find them. Asking the AI to create a YouTube search URL solves this. If you state your desired topic and conditions, the AI uses advanced YouTube search syntax to create an accurate search address.

YouTube Search URL Creation Prompt Example 1 - Finding English Academic Seminars.

Please create a YouTube search URL that meets the following conditions: - Topic: artificial intelligence ethics seminar - Video length: 20 minutes or more (excluding short clips) - Upload time: Within the last year - Channel type: University, research institute, academic society official channels first - Sort by: View count or relevance.

YouTube Search URL Creation Prompt Example 2 - Finding Multilingual Seminars. Please create YouTube search URLs for the following topics in each language: - Topic: Future of Generative AI - English: generative AI future conference - Japanese: 生成AI 未来 セミナー - German: generative KI Zukunft Konferenz - French: IA générative avenir conférence.

Please create separate search URLs for each language. YouTube Search URL Creation Prompt Example 3 - Finding Presentations from Specific Organizations.

Please create a YouTube search URL that meets the following conditions: - Channel: World Economic Forum, TED, MIT, Stanford official channels - Topic: AI regulation policy - Period: Uploaded after 2024 - Video length: 15 minutes or more. There are two ways to use the seminar videos found through these searches in NotebookLM.

The first method is to put the YouTube link directly into NotebookLM. NotebookLM accepts YouTube videos as a source and automatically reads and analyzes the video's transcript. It can handle transcripts in various languages including English, Japanese, and Korean.

The second method is to first extract the script of the video using AI. If you click 'More' under the YouTube video, there is a 'Show Transcript' button. Copy this and save it as a text file.

Alternatively, you can ask the AI, 'Please organize the entire script of this YouTube video. ' Seminars in foreign languages are no problem; AI has excellent translation capabilities, so it can even summarize presentations conducted in Japanese or German into English. You can grasp the latest research trends worldwide without language barriers.

Seminar Script Organization Prompt Example: Please organize the contents of the following YouTube video: [YouTube Link]

1. Speaker's Name and Affiliation

2. Three Key Arguments of the Presentation

3. Evidence or Data Supporting the Arguments

4. Future Outlook Mentioned in the Presentation

5. Important Content from the Q&A Session

If the original language is English, Japanese, or German, please translate and organize the content into Korean. You can gain richer insights by uploading multiple seminar videos to NotebookLM together.

For example, if you input four AI policy seminars held in the US, Europe, Japan, and Korea and ask, "Compare how the AI regulatory approaches of each region differ," you can grasp global trends at a glance.

Example of a Comprehensive Multilingual Seminar Analysis Prompt in NotebookLM: Please conduct a comparative analysis of the four uploaded seminar videos (USA, EU, Japan, Korea):

1. What is the issue of greatest concern regarding AI in each region?

2. Organize differences in regulatory approaches into a table (Strong Regulation vs. Self-Regulation)

3. What principles, if any, are commonly agreed upon?

4. Differences in the solutions proposed by each speaker

5. Areas mentioned as requiring future international cooperation

Please indicate the source of each piece of content, specifying which seminar it came from. The greatest advantage of this method is "time-saving." Watching five one-hour seminar videos takes five hours.

However, by putting them into NotebookLM and extracting only the core points, you can grasp the entire content in 30 minutes. Of course, it is better to watch important videos directly. But it is much more efficient to identify which videos are important first and then watch them selectively.

C. Practical Workflow: From Search to Insight. Let's organize the content explained so far into a single flow.

For example, suppose you are writing a report on the topic of "2025 Global AI Regulation Trends." Step 1: Ask AI to Create Search URLs.

Please create search URLs where the following materials can be found: [For Google Search] - Topic: AI regulation 2024 2025 - File: PDF - Sources: European Commission, White House, OECD, Korean Government - Period: Since 2024.

[For YouTube Search] - Topic: AI Act EU implementation / AI regulation summit - Length: 20 minutes or more - Period: Past 6 months - Channel: Official institutional channels. Step 2: Data Collection • Access the Google search URLs and download 5–10 PDF reports. • Access the YouTube search URLs and select 3–5 key seminar videos.

• Copy the video links or save the scripts as text files. Step 3: Upload to NotebookLM • Create a new notebook: "2025 Global AI Regulation Trends." • Upload all collected PDF files and YouTube links.

• A summary is automatically generated once the upload is complete. Step 4: Extract Insights with Step-by-Step Questions. [1st Question - Overall Understanding] Please summarize the five core principles of AI regulation commonly mentioned in all uploaded materials.

[2nd Question - Comparative Analysis] Please compare the differences in approach between the EU AI Act and the US AI Executive Order in a table. Comparison items: subject of regulation, risk classification system, penalty levels, enforcement timing.

[3rd Question - Trend Discovery] When comparing the 2024 and 2025 data, please find any newly emerging issues or changes in tone. [4th Question - Identifying Research Gaps] Please organize items mentioned in the materials as "unresolved issues" or "areas requiring further research."

[5th Question - Deriving Actionable Proposals] Please summarize three overseas cases and policy proposals that the Korean government could refer to. Step 5: Organize Results • Save NotebookLM's answers in notes. • Click on citation numbers to verify the original text.

• Use the Audio Overview feature to review the entire content by listening again. • Visualize the relationships between materials with a mind map. By following this workflow, you can quickly collect the latest official documents from around the world and perform in-depth analysis with the help of AI.

Tasks that used to take a researcher several weeks can now be completed in a single day. You can spend the remaining time interpreting data and developing your own thoughts.

4. Paper Structure Design

A research paper is not something you write randomly; it follows a set order and format. Just as a blueprint is needed when building a house, you must first establish the skeleton when writing a paper. This is called "Structure Design."

A solid structure makes the same content more persuasive, whereas a loose structure fails to convey even the best research results to the reader. Most academic papers follow the IMRaD structure: I for Introduction, M for Methods, R for Results, and D for Discussion. The Introduction explains why the research is necessary.

Methods describe how the experiment or survey was conducted. Results show what was discovered. Discussion interprets what those results mean.

Using AI can significantly reduce the time spent on structure design. However, simply asking AI to "make a table of contents for a paper" without a plan won't yield good results. You must first summarize the core of your research into a single sentence.

A format like "How A affects B" is easy to work with—for example, "How does smartphone usage time affect the sleep quality of adolescents?" Paper Structure Design Prompt Example 1 - Requesting a Table of Contents (TOC) Draft Research Question: How does smartphone usage time affect the sleep quality of adolescents?

Research Method: Survey (Target: 200 middle school students) Instruction: Please create a table of contents in IMRaD format suitable for the research above. Explain in 2–3 lines what content should be included in each chapter.

Paper Structure Design Prompt Example 2 - Framing the Introduction Please create an outline for the introduction in the following order:

1. Problem Situation (3 sentences): Current status of excessive smartphone use among adolescents

2. Limitations of Existing Research (2 sentences): What remains unknown

3. Research Question (1 sentence): The question this paper aims to answer

4. Research Contribution (2 sentences): In what way is this research new?

Paper Structure Design Prompt Example 3 - Comparing Multiple Structural Proposals My Research Topic: The impact of online games on adolescent friendship formation.

Please provide the following three structural proposals: - Proposal 1: Traditional IMRaD format - Proposal 2: Problem-Solution format - Proposal 3: Pros and Cons comparison format Also, explain the strengths and weaknesses of each structure. The Introduction is not merely a section for explaining background; it is the first gateway to persuading the reader.

It must answer the questions: "Why is this research needed?" , "What has been lacking so far?" , and "What new insights does my research reveal?"

When asking AI to write an introduction, instruct it to compose paragraphs in the order of answering these three questions. The Methods section is written for Reproducibility. Others should be able to achieve the same results if they follow your research process exactly.

Therefore, you must record every detail: how the questionnaire was developed, how many people were surveyed, and how the responses were analyzed. Results and Discussion are distinct. Results only state objective facts.

For example, you show numbers and data like "65% of survey respondents answered that they use a smartphone for more than 3 hours a day." Discussion interprets what those results mean. You might write, "This result indicates high smartphone dependency among adolescents, which could be a cause of sleep deprivation."

If you instruct the AI to "keep observations in Results and interpretations in Discussion," the two parts will not get mixed up. There is also a routine for using NotebookLM for structure design. First, upload 5–10 reference papers to NotebookLM.

Then, ask it to "summarize the table of contents structure commonly used by these papers." Referencing the structures of high-quality papers will help make your own paper's skeleton more robust.

5. Insight Extraction Methodology

"Insight" means "deep understanding." It refers to finding the hidden meaning beyond surface-level information. For example, stating that "70% of respondents said they sleep poorly" is a fact.

The statement "Since most of them use smartphones after 11 PM, it appears that smartphone blue light has interfered with sleep hormone secretion" is an insight. It goes one step further from the facts to answer "Why is it so?" and "So what?"

Insight extraction is the core process that transforms a mundane report into valuable research. Anyone can summarize data, but true research involves finding meanings that others have missed. Putting your own experimental data into NotebookLM makes insight extraction much easier.

Information already on the internet is available to everyone, but the data you measured, your survey results, and your observations are unique. AI finds patterns in this original data. Insight Extraction Prompt Example 1 - Pattern Discovery Please find 5 recurring patterns in my experimental data.

Provide specific numerical evidence for each pattern. If there are relationships connecting the patterns, explain those as well. Insight Extraction Prompt Example 2 - Finding Opposing Evidence My research results support the claim that "smartphone use degrades sleep quality."

Find any content among the uploaded materials that contradicts this claim. If there are opposing views, compare how they differ from my research. Insight Extraction Prompt Example 3 - Finding Research Gaps Please summarize the common limitations mentioned in the 10 uploaded papers.

Suggest 5 Research Gaps that have not yet been studied. Explain why each gap is important for future research. Comparison is a very powerful method for finding insights.

Place A and B side-by-side to find differences. Consider questions like "How much does the growth rate differ between a group of plants played music and a group that was not?" or "How do smartphone usage policies for adolescents differ between Korea and Japan?"

Asking AI to create a comparison table makes these differences clear at a glance. However, be cautious. AI tends to use hyperbolic expressions.

You should be skeptical if words like "innovative," "completely new," or "decisive discovery" appear. Good insights consist of accurate words, not "big" words. Summarizing your findings into three sentences—"Observation, Evidence, Limitation"—removes exaggeration.

For example: • Observation: The group of plants played music grew an average of 2cm more. • Evidence: Consistent differences appeared in the data measured daily for 30 days. • Limitation: It is difficult to generalize due to the short experiment period and small sample size.

Organizing in this way leads to honest insights without exaggeration.

6. Ethical Use Standards for AI Assistants in Academic Writing

While AI is a convenient tool, it should not be used carelessly. In writing where you are being evaluated, such as school assignments or research papers, Ethics are extremely important. Ethics refers to the duties and principles one must rightfully follow.

There are rules to observe when using AI as well. The most important principle is that "AI is a tool, not an author." Even if AI helps write the text, the owner of that writing is you.

You also bear responsibility for the content. The excuse "The AI did it" does not work. Just as the person behind the wheel is responsible for an accident, the person who submits the writing is responsible for it.

Transparency is also vital. If you received help from AI, you must state it honestly. You should write at the end of the paper or in the acknowledgments: "ChatGPT was used in the literature review process of this paper."

Trying to hide it can lead to problems. Nowadays, there are programs that detect AI-written text, making it easy to get caught. Stating it honestly allows teachers to recognize that you have "used tools smartly."

AI usage policies vary by academic journal. The Nature Publishing Group requires the use of AI tools to be explicitly stated in papers. You must specify which AI was used and how.

Science journal is even stricter; you cannot use AI-generated text without editor permission. When doing school assignments, it is best to ask your teacher first: "Can I use AI for this assignment? If so, to what extent is it allowed?"

Ethical AI Use Checklist □ Have I avoided copying and pasting AI-written content as-is? □ Have I double-checked the information provided by AI with other sources? □ Have I explicitly stated the fact of AI usage in the paper or assignment?

□ Am I aware that the responsibility for the final result lies with me? □ Have I avoided inputting personal information or sensitive data into the AI? You must distinguish between permitted and prohibited uses.

Brainstorming ideas, grammar correction, translation drafts, and data summarization are generally permitted. However, submitting AI-written text as your own, manipulating experimental results, or writing AI text without attribution is prohibited. You must also know AI's limitations.

AI can have Bias. If the training data contains many specific perspectives, the results will lean that way. It can also fabricate facts through Hallucination.

Do not trust AI's answers unconditionally; always review them critically.

7. Plagiarism Prevention Principles

Plagiarism is secretly taking someone else's thoughts or writings and using them as if they were your own. In school, it means copying a friend's homework; in research, it means stealing another researcher's paper. Plagiarism is essentially theft.

In school, you get a failing grade; for researchers, it deals a major blow to their career. In the AI era, plagiarism seems easier. If you ask AI to "write a book report," the text comes out quickly.

However, submitting this under your own name as is constitutes plagiarism. What AI writes is not what you thought; it is the same as copying someone else's ideas. The most reliable way to prevent plagiarism is to go through three steps: "Original Source - My Notes - My Sentence."

Step 1: Highlight important content in the original text. Step 2: Take notes on that content in your own words. Do not copy the original verbatim.

Step 3: Based on your notes, write new sentences that fit your logic. Through this process, you digest someone else's thoughts as your own. You must also include citations.

If you borrowed someone else's idea, you must state the source. Indicate it by stating, "This content was taken from Professor Chul-su Kim's 2024 paper." Plagiarism Prevention Routine

1. Write down citation information immediately when reading materials (author, year, title, page)

2. Check every sentence when writing a paper: "Is this my argument, or someone else's evidence?"

3. When using someone else's sentences verbatim, use quotation marks ("") and indicate the source

4. Run a plagiarism check tool (Turnitin, CopyKiller, etc.) before submission

5. Rewrite sections that show high similarity

There are cases where the risk of plagiarism increases when using AI. If you give AI a paragraph from a paper and ask it to "polish it up," sentences similar to the original may remain. Changing only a few words is called patchwriting and is still considered plagiarism.

To avoid this, you should tell AI, "Don't look at the original text; write a new sentence based only on the notes I provided." The bibliography is also part of plagiarism prevention. Using references generated by AI as they are is dangerous.

Non-existent papers might be mixed in. Be sure to include only "materials I have personally opened and verified" in the bibliography. Check the DOI (Digital Object Identifier), journal name, and publication year before writing them down.

In conclusion, while AI speeds up writing, it can also speed up plagiarism. As speed increases, rules become more important. AI must be used honestly, transparently, and responsibly.

That is the most important virtue we must possess when writing in the AI era.