Tutorial: Writing Text with LLM

Avoid and Detect Bullshit

Jan Lorenz

Text generating AI Tool

• Using text generating AI tools has become a reality and is used for
  • learning,
  • translating,
  • coding,
  • receiving advice,
  • leisure, and
    
  • for writing and revising text for us.
• The backbone of the generative AI tools ar large language models which generate text

Use as a student

• Help to solve assessment and other tasks ☑️
  👍 Especially for coding, but understand and scrutinize output!
  ⚠️ Don’t forget the learning!
• Learning by asking for explanations ✅
  Adjust the detail level 📑 and specificity 🔎 with follow up prompts.
  ⚠️ Don’t stop at the chat, get productive 🧑‍💻, find authorative sources and continue 📚.
• Generate ideas for your own project (☑️)
  ⚠️ Often generic unspecific ideas.
  🧑‍💻 quickly start to get productive.
• Write text for your own project reports ((☑️))
  ⚠️ Often generic unspecific text.
  ✏️ You have to write the core research questions, arguments, and results yourself!

Prompting workflow

Text development works with chatting via a chain of prompts

1. You write what you want
   Write an outline for data science project about cricket.
2. You ask for refinement, reshaping, and restructuring on further prompts
   Write an abstract of about 200 words on it.
3. Continue until you are satisfied. Copy, edit and finalize yourself.

How LLMs work: Autocomplete on Overdrive

Activity

Follow The Bullshit Machines

Chapter 1: Autocomplete on Overdrive

Questions:

• Does the same prompt always lead to the same outcome?
• (From The Bullshit Machines) Why do you think a program that simply tries to predict the most likely next word in a sequence would be good at holding conversations?

How LLMs work: Autocomplete on Overdrive

Activity

Why do some call LLM Bullshit Machines?

Follow The Bullshit Machines

Chapter 2: The Nature of Bullshit

Questions:

• (From The Bullshit Machines) People are perfectly good at producing bullshit without AI assistance—but with AI, people can produce more bullshit, faster. Who might find that useful? How?

The instructors perspective

The instructor

• has to read a lot
• must read all student reports with an equal grading mindset
• should provide some constructive feedback

Problem

AI generated text increases the amount of text which reads (superficially) well but it can become even harder to find the own contribution. It may read a bit like bullshit. Often it is very generic.

AI may be used also by the instructor!

• Summarize student contributions
• Write Feedback

(I haven’t done so far ✌🤞️.) However, we want to avoid this

Is using text generating AI plagiarism?

Activity

Two (or more) groups to answer the same questions:

• What is plagiarism?
• Is using AI to write text a form of plagiarism? Why? Why not?

Group 1 Read Constructor University’s Code of Academic Conduct

Group 2-X Ask AI!

Data Science Project Abstract Writing

Activity

Let us try different AI Tools: OpenAI’s ChatGPT, Antrophic’s Claude, Mistral AI’s Le Chat, Google’s Gemini, …

Group 1-X Two on ChatGPT, more with other tools

Use this prompts:

1. You write what you want
   Write an outline for data science project about cricket.
2. You ask for refinement, reshaping, and restructuring on further prompts
   Write an abstract of about 150 words on it.

Discussion: Read out the texts and we discuss its quality from the educated reader’s and instructor’s perspective.

Final Activity: Abstract for your project

Activity

In your group:

Create a project abstract of about 150 words for your project with as many prompts as you want.

(10-15 minutes)

Discussion: Read your abstracts and we discuss its quality from the educated reader’s and instructor’s perspective.