My idea is to have one generic single prompt for a Large Language Model like ChatGPT to:
Read an article or blog post.
Generate a list of atomic ideas to select from.
Create Literature Notes from selection in a given structure and format.
PROMPT: Read article and create notes. v1.5
Hello my friend,
Thank you for taking the time to talk to me. Here is my prompt framework with five steps I want to use to initiate our conversation.
Hints: The prompt framework is formatted in mark down. Do not use text marked as comment (``%%text%%``).
***
**Step 1: Setting the Stage**
- Act as an experienced user of Zettelkasten methodology from Niklas Luhmann as published by Sƶnke Ahrens.
**Step 2: Defining the Objective**
- I want you to support me in generating Literature Notes from a given article or blog post.
**Step 3: Structuring the Interaction**
1. **Start Interaction:**
Please ask me for getting an article for further investigation.
2. **Extract Atomic Ideas:**
Please analyze the given article or post and identify its fundamental atomic ideas. These are the smallest, indivisible concepts or pieces of information present in the post. List them out for further processing.
3. **Identify Key Insights:**
From the list of atomic ideas you've extracted, identify the most important and significant insights. These are the main takeaways or concepts that the blog post is conveying.
List these key insights in a numbered list.
Use a grading scale with a range from 1/10 to 10/10. Use 1/10 for least important, 10/10 for most important.
Sort key insights from high to low.
4. **Ask for selection**
Please ask me to select some of these insights for further use.
5. **Create Summaries:**
For each key insight Iāve selected, write a concise summary in a paragraph. The summary should capture the essence of the insight while remaining succinct and informative.
6. **Generate Descriptive Titles:**
Using the key insights and summaries you've created, come up with descriptive titles for each of these insights. These titles should provide a clear idea of what the insight is about, while also being catchy and engaging. Titles are no longer than 10 words. They only use characters and numbers, no special characters.
**Step 4: Requesting Specific Information**
- Whenever you generate an answer explain the reasoning and assumptions behind your answer so that I can improve my questions.
**Step 5: Summarizing and Concluding**
- Please prepare your results in mark down format and use the following template:
- - -
**TITLE**
> SUMMARY
Tags: LIST OF TAGS
Keywords: LIST OF KEYWORDS
- - -
- Do not use `:` in TITLES.
- Words in CAPITAL LETTERS are placeholders for your input. Start with a new line for each item.
- Create LIST OF TAGS with nested tags like: ``type/note`` ``theme/zettelkasten`` etc.
***
From now on use the rules above. But before starting, feel free to ask me any question that may help you to improve the quality of your answers.
%%
PROMPT - Read article and create notes, v1.5, 2023-08-30
by Edmund Grƶpl
More about: https://github.com/groepl/Obsidian-Templates
%%
Iām sure it will need some experiments to get best results. What are your ideas?
Iād say you should always write your own notes, not have AI do it for you, because writing is thinking and writing the notes forces you to understand it more deeply than copying down AI content.
Iāve also been experimenting with AI for this purpose, and I still am, with part of the focus on what prompt to give it, but more importantly how exactly does it fit into my workflow.
Identifiying key takeaways or key insights like you have in your prompt seems like a harmless one to me that doesnāt interfere with my understanding of the subject matter. That being said, I also want to be able to connect my own thoughts and ideas, so Iām considering adding the ones I select from it under a special header which I call AI thoughts. Weāll see.
Congrats on working on this prompt so much. Iāll test it out.
I especially like how you started it, since I try to use AI like that as well in this process: using it as a friend that would be interested to talk about that topic and help me identify or process my thoughts. Almost like a brainstorming partner.
Yes for sure. But sometimes itās a good training for my brain to compare the result from AI with my own thinking. For me AI is one of my preferred dialogue partners.
Talking to a friend makes a difference. And ChatGPT already knows the concept of being a friend. We can also use the āPersona Promptā at the beginning of our dialogue: āAct like a friend.ā
I find itās sometimes useful to ask LLMs to explain a difficult concept in plain English, then summarize the response by hand or through another prompt to record in my commonplace notes.
Yes, if you donāt get something, you shouldnāt just copy & paste. However, if youāve already got it and want to have a quick refresher in the future when you inevitably forget, then itās fine to copy an LLMās response IMO.
āSummarize the above response into a shorter summary that could perceivably fit onto an index cardā is a particularly useful prompt.
As a software developer, Iād be intrigued in using a chatbot the same way I use github copilot. Over there, AI isnāt reallly asked questions, it just analyzes your code and acts as a better autosuggest based on what youāve just written.
That can be nice and would be something Iād be trying out. Other than that, using it as a conversation partner after youāve already written your notes down and you ask them to see whether they bring up new aspects you hadnāt considered yet seems fine-
Some of us use large language models to improve their writing from Zettelkasten, others use them to support thinking. Basically, we always have four options. Here are some of my preferred examples:
1 - Spell Checking
Prompt: Act like an editor. Please check the following note from my Zettelkasten for spelling and grammar errors.
2 - Idea Generation
Prompt: Your are an expert in <theme>. Please suggest <n>outstanding ideas related to <theme, topic> for an audience like <audience example>.
3 - Socratic Dialog
Prompt 1: Help me to understand <theme, topic>. Initiate and lead a dialogue with me by asking socratic questions. Expect my answers and react with a new question.
Prompt 2: Please give me an overview on the key insights from this session.
Prompt 3: Use this summary to write a newspaper article with headline, lead and text. Prepare the article in a style like <named style>.
4 - Article Writing
Prompt: Please write a newspaper article about <theme, topic> with headline, lead and text. Prepare the article in a <preferred style>.
Hello sir, where can i ask a questions that are about your workflow for example your folders overview? I am struggling with this a lot. I figured out from this picture (1) that the folders should be something like this (2)
For me, blindly submitting content for summary bullet points has not been a useful endeavor. In comparing the ai bullet points with my own after reading the content it became obvious that my summary was way better and, not to mention, prompting chews up api time with the llm; itās costly. Most āinboxā content I scan and summarize on my own. That capture dopamine thing is such a seductive beast:) I just say no to this kind of prompting.
Another observation - determining the salient bullet points is completely different when submitting content and the same prompt to gemini, claude, and chatgpt - whoās bullets you gonna believe? That said, the more you invest in the setup for the prompt, the better your output; the ai has no idea who you are or what you want. Sorta garbage in, garbage out. In the end, read and think about the content; your own bullet points will always be better.
Content summarizing, spell checking, email crafting, and article writing - not a fan, to me, thatās just wrong. We should be fluent and do these without thinking about them; theyāre basic human communication skills - use them or loose them.
Where I do find value in ai prompting is the terms @Edmund offered up in this thread: idea generation and socratic dialog. (I just queried claude - what is socratic dialog).
In a novel Iām writing, I was challenged to understand hydrology in basin and range terrain; particularly toxic closed basins. Short of reading all four volumes of Mcpheeās āAnnals of the Former Worldā, I sat with claude and interatively prompted to gain a good understanding; ai is a phenomenal learning tool, if used properly.
With regard to the socratic thing: My niece is a member of a interdisciplinary team of graduate students who were hired to analyze a agricultural co-op in transition due to climate change. The landowner was keen on regenerating the farm, moving away from range fed beef cattle to rotating pastured sheep and implementing watershed restoration to grow plant crops, while still maintaining and safeguarding the co-op families that had been a fundamental part of the operation for generations. This meant new livestock and plantings which implied new processes, products and challenges for the co-op; personal and social change. The team was tasked with analyzing the project but couldnāt socratically come up with a plan. After several phone calls with my niece, we came up with a setup and series of prompts that really got the ball rolling for them. She says they modify it all the time to drill down into what they want.
@ryadaj offers excellent setup and prompting ideas in the link mentioned above!
@edmund - I always appreciate your posts. They are in depth and intended for thought. That sort of depth and focus is rare these days. Cheers!
