An image of PAT (Paper Authoring Tutor), generated by Stable Diffusion XL v1.0 A Guide to Paper Writing Support with Generative AI A Joint Zemi Kenji Saito, Hiroshi Kanno (Professors, Waseda Business School) and friends A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.1/23

The slides can be found at : https://speakerdeck.com/ks91 In addition, you are invited to the Joint Zemi Discord You will see the URL and QR code The Discord server offers unlimited use of tools such as PAT (Paper Authoring Tutor) and The (multi-lingual) Interpreter, and you can even create your own AI assistant (based on OpenAI API) if you want A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.2/23

In This Joint Zemi, As you are all aware of the recent development of so-called “generative AI,” Waseda University is seeking to utilize this technology under its basic policy of training and nurturing “robust intellect and flexible sensibility” We believe that WBS students who are preparing to write their thesis need to know how to use the AI effectively and beneficially, and as graduate students who are independent scholars, you need a place to discuss how to use the AI yourselves So, we have planned this joint Zemi to explain the simple principles, features, and limitations of generative AI, to introduce appropriate and advanced uses related to research and paper writing, to answer your questions so that you can use this new tool with confidence, and to discuss how to make use of it with everyone A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.3/23

Today’s Agenda What is a research paper anyway? Demonstration Types and Overview of AI Tools Finding Papers Using AI Reading with AI Analyzing with AI Writing with AI Discussion on how to make use of AI (if there is time) A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.4/23

What is a research paper anyway? It is there to expand the knowledge of humankind as we all check to see if our new knowledge is correct Paper has a claim (new knowledge) Primary data is collected by the authors themselves to support the claim We explain the gap with the secondary data, which is the knowledge to date The above also applies to business school papers New quantitative/qualitative research, business proposal, new findings from surveys of prior cases/studies, or a combination of these A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.5/23

Research and Publications — So far Discoveries and inventions by humans The entire scope of human knowledge paper paper We cannot read them all already Driven by incentives such as degrees, prestige, and curiosity ɹɹɹ So things like who wrote it, who was the first, who stole ideas, etc. are important! The entire scope of knowledge is expanded by checking (through peer review and post-publication discussion) whether the new knowledge is correct A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.6/23

Demonstration If you have already started your research, you must have presented your work in progress in zemi or at least created some kind of materials We will demonstrate how to feed such materials into the writing mode of our PAT (Paper Authoring Tutor) and output a draft of the paper As just one of the possibilities of what can be done A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.7/23

About PAT (Paper Authoring Tutor) An AI assistant to assist you in conducting research and writing papers Can be found at : GPTs : https://chatgpt.com/g/g-6lW16zlHo-hatuto-pat-lun-wen-asisutanto Also in the joint zemi Discord server GitHub : https://github.com/ks91/pat Researchers’ manual (in progress) : https://speakerdeck.com/ks91/pat-paper-authoring-tutor-researchers-manual You can see what’s in it (see GitHub), so you can effectively ask it questions to make it work properly, or you can improve it yourself and register it in GPTs, etc. Can be ported to large language models other than GPT A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.8/23

Demonstration Upload this presentation slides https://speakerdeck.com/ks91/simulation-of-economic-activities-related-to-space-debris-using-a-serious-game “This is a slide file from a presentation I gave at a recent JAXA workshop. I would like to finalize this content into a research study, and to do so, I would like to tentatively consider it in the form of a paper. Could you draft the paper in markdown format, using what you can extract from this slide to fill in the appropriate answers to the 10 questions in the writing mode?” Next, let’s have it in L A TEX format, and then we’ll try to view it in Overleaf as a rough draft If you have your own materials, please do the same We encourage you to try everything in the workshop out for yourself A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.9/23

Types and Overview of AI Tools This time we will focus on the application of large language models There are many more than what is listed here, and new ones are coming out all the time, so let’s share what we know! What is a large language model anyway? A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.10/23

What is a Large Language Model? Large version of a language model What is language model? A statistical model that says that in this language, words are often arranged in this way If you give it words up to the middle of a sentence, it will generate the common content that follows it (in an appropriately random manner) Applying it to conversation, you get ChatGPT for example Therefore, When we tell them to “take a deep breath and think it over,” they respond with something that sounds like they’ve thought it over What is generated looks very much like that, but it is often not true, even if it is correct in English, especially when it comes to content that does not appear that much on the Internet When researchers ask technical questions, they often answer incorrectly Giving them lots of context makes it easier to get good results A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.11/23

What is important is that . . . YOU are responsible for the content of your paper! YOU do the fact checking! Protect yourself and your new knowledge A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.12/23

AI Tools for Research (there should be more) Finding papers Perplexity (Perplexity AI) Reading NotebookLM (Google), Specialized GPT such as PAT, etc. Analysis Claude (Anthropic), ChatGPT (OpenAI)(w/ Code Interpreter), Open Interpreter (Tool to write and run Python and/or R code) Writing Claude, ChatGPT, Specialized GPT such as PAT, etc. A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.13/23

Finding Papers Using AI What do you search for in the first place? Look for secondary data that may have gaps with your claims Papers are also data You may have to explore what your claims are through this process A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.14/23

Finding Papers Basically, search within the past 3 years using Google Scholar, etc. Semantic Scholar is also useful Why within 3 years? → Because we look for gaps with the latest knowledge How about Perplexity? Convenient, but it can go wrong (You may search for papers on Perplexity and then search again on Google Scholar for the ones you find) If it’s your area of expertise, you can realize, “oh it’s wrong,” and give Perplexity the right knowledge to redo it, but if not, it could be dangerous If the papers/books you search are to be used as secondary data, you must know exactly what they are If you get interested, get the text first And then read it with the help of AI A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.15/23

Reading with AI NotebookLM as an example NotebookLM, like PAT, is a research assistant Very useful for quickly grasping the literature A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.16/23

Reading Papers If you throw the data into NotebookLM, it will first summarize it, so you can determine if it is appropriate as secondary data In the past, when we found a paper, we would first identify it by its abstract and conclusion, but the above is an alternative to that You can ask questions about the content and get answers, but read the relevant sections yourself to see for yourself While it may be tempting to use the text as written by LM, it is always better to quote from the text to ensure accuracy and avoid unintentional plagiarism When reading critically, one single source is basically better You can avoid mixing other knowledge in the results Uploading multiple sources provides emergent answers That is one way to use it A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.17/23

Have Your Papers Read Once you have written some of your own papers, upload them to NotebookLM or PAT for feedback You can have them point out strengths and weaknesses, and Where things do not add up logically Peer review could also be simulated A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.18/23

Analyzing with AI If you do not have the knowledge, you cannot control the analysis process Knowledge from “Business Data Analysis” should help A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.19/23

Analysis with Open Interpreter Multiple comparisons of the following simple data sample1 <- c(3.10, 3.14, 3.07, 3.20, 2.84) sample2 <- c(2.76, 2.88, 2.88, 3.08, 2.93) sample3 <- c(3.19, 3.13, 3.45, 3.34) sample4 <- c(2.84, 2.72, 2.61, 2.65, 2.61) “In the current directory, there is a dataset called "data.txt" written in R vector format, consisting of four samples, which should be compared multiply. Also, draw a box-and-whisker diagram with alphabetical letters indicating significant differences so that human beings can visually compare the results.” A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.20/23

Writing with AI Once again : you are responsible for the content of your paper A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.21/23

Get help writing your paper It is convenient to have the whole thing composed Originality is not required in the composition of the paper Where the writing is tedious and would be the same no matter who wrote it, it would be helpful to have the AI draft it for you Upload a diagram and have it explained Upload a graph and ask it to write down what it can interpret from it, etc. Shorten without sacrificing content is also often used It may not be necessary for a master’s thesis (or equivalent)! If you are submitting a paper to a conference or journal, there may be a page limit “Summarize” would lose information and would not be usable as the body of the paper A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.22/23

Discussion on how to make use of AI A Guide to Paper Writing Support with Generative AI — 2024-12-17 – p.23/23