AI Assisted Knowledge Consumption

> [!danger] The Problem > Distilled efficient information delivery is difficult. As such, most content is incredibly inefficient and redundant. Paired with [[The Memory Problem]], a lot of time is wasted. > > One solution is to use AI to summarise information. However, basic summarisation prompts are ineffective as they result in catastrophic losses of information. > [!check] Solution > Few-shot prompt engineering. **Results:** [[Optimal Protocols for Studying & Learning (Huberman Lab)#Index|Podcast Example]], [[Why Tacit Knowledge is More Important Than Deliberate Practice (Cedric Chin 2021)#Index|Article Example]], [[PSYC30016 Lecture 7#Index|University Lecture Example]] # Step I ## Written Article ``` Create an EXHAUSTIVE index of the given text. This task is CRITICAL and requires ABSOLUTE PERFECTION. Any omission, no matter how small, results in complete failure. You may use the attached lecture slides to help. Instructions: 1. Index EVERY SINGLE sentence, phrase, or distinct idea in the text. 2. Use a hierarchical structure (1, 1.1, 1.2, 2, 2.1, etc.). ALWAYS MIMINC THE TEXT'S STRUCTURE AND THEN EXPAND. 3. Each index entry must use keywords, describing ONLY the type of information present, NOT the information itself. 4. Include ALL mentions of concepts, examples, definitions, or explanations. 5. Note ALL transitions, introductions, and conclusions. 6. Indicate ALL visual cues or directions mentioned in the script. Critical Reminders: - COMPLETENESS IS PARAMOUNT. Do NOT skip ANY detail, no matter how insignificant it may seem. - If you're tempted to omit something, STOP and index it anyway. - Your index should enable someone to reconstruct the entire structure of the video without watching it. - Failure to capture ANY element of the script means the entire task has failed. - There is no length constraints, continue until completion. Example format (make sure to only go max 4 layers, output in markdown format with indentations, P represents paragraph number, L represents line number): 1. Introduction (P1L1 - PxLx) 1.1. Sub topic 1 (P1L1 - PxLx) 1.1.1. Sub sub topic (PxLx - PxLx) 1.1.1.a. Sub sub sub topic (PxLx - PxLx) 2.Topic (PxLx - PxLx) 2.1. Sub topic 1 (PxLx - PxLx) 2.1.1. Sub sub topic (PxLx - PxLx) 2.1.1.a. Sub sub sub topic (PxLx - PxLx) ``` ``` Create an EXHAUSTIVE index of the given written passage or book paragraph. This task is CRITICAL and requires ABSOLUTE PERFECTION. Any omission, no matter how small, results in complete failure. You may use any accompanying notes or resources to help. Instructions: 1. Index EVERY SINGLE sentence, phrase, or distinct idea in the passage. 2. Use a hierarchical structure (1, 1.1, 1.2, 2, 2.1, etc.) with specific subheadings for each paragraph or major section. 3. Each index entry must use keywords, describing ONLY the type of information present, NOT the information itself. 4. Include ALL mentions of concepts, examples, definitions, or explanations. 5. Note ALL transitions, introductions, and conclusions within each paragraph. 6. Indicate ALL references, quotes, or citations mentioned in the passage. Critical Reminders: - COMPLETENESS IS PARAMOUNT. Do NOT skip ANY detail, no matter how insignificant it may seem. - If you're tempted to omit something, STOP and index it anyway. - Your index should enable someone to reconstruct the entire structure and content of the passage without reading it. - Failure to capture ANY element of the passage means the entire task has failed. Example format: - 1. Paragraph Topic Heading - 1.1 Subtopic/Key Idea 1 - 1.1.1 Sub-subtopic - 1.2 Subtopic/Key Idea 2 - 2. Next Paragraph Topic Heading - 2.1 Subtopic/Key Idea 1 - 2.1.1 Sub-subtopic - 2.2 Subtopic/Key Idea 2 Output Format: Provide the index in markdown format, ensuring the structure is clear and easy to navigate. Each paragraph should have a clear topic subheading, followed by a detailed breakdown of the content.IMPORTANT: The numbered topic heading are only bolded and are dotpointed, they are not formatted as headings. ``` ## Video Script ``` Create an EXHAUSTIVE index of the given video script. This task is CRITICAL and requires ABSOLUTE PERFECTION. Any omission, no matter how small, results in complete failure. You may use the attached lecture slides to help. Instructions: 1. Index EVERY SINGLE sentence, phrase, or distinct idea in the script. 2. Use a hierarchical structure (1, 1.1, 1.2, 2, 2.1, etc.). 3. Each index entry must use keywords, describing ONLY the type of information present, NOT the information itself. 4. Include ALL mentions of concepts, examples, definitions, or explanations. 5. Note ALL transitions, introductions, and conclusions. 6. Indicate ALL visual cues or directions mentioned in the script. Critical Reminders: - COMPLETENESS IS PARAMOUNT. Do NOT skip ANY detail, no matter how insignificant it may seem. - If you're tempted to omit something, STOP and index it anyway. - Your index should enable someone to reconstruct the entire structure of the video without watching it. - Failure to capture ANY element of the script means the entire task has failed. Example format (make sure to only go max 4 layers, output in markdown format with indentations): - 1. **Introduction** (00:00 - xx:xx) - 1.1 **Sub topic 1** (00:00 - xx:xx) - 1.1.1 **Sub sub topic** (xx:xx - xx:xx) - 1.1.1.a **Sub sub sub topic** (xx:xx - xx:xx) - 2. **Topic** (Lx - Lx) - 2.1 **Sub topic 1** (xx:xx - xx:xx) - 2.1.1 **Sub sub topic** (xx:xx - xx:xx) - 2.1.1.a **Sub sub sub topic** (xx:xx - xx:xx) ``` ### Uni Format ``` Create an EXHAUSTIVE index of the given video script. This task is CRITICAL and requires ABSOLUTE PERFECTION. Any omission, no matter how small, results in complete failure. You may use the attached lecture slides to help. Instructions: 1. Index EVERY SINGLE sentence, phrase, or distinct idea in the script. 2. Use a hierarchical structure (1, 1.1, 1.2, 2, 2.1, etc.). 3. Each index entry must use keywords, describing ONLY the type of information present, NOT the information itself. 4. Include ALL mentions of concepts, examples, definitions, or explanations. 5. Note ALL transitions, introductions, and conclusions. 6. Indicate ALL visual cues or directions mentioned in the script. Critical Reminders: - COMPLETENESS IS PARAMOUNT. Do NOT skip ANY detail, no matter how insignificant it may seem. - If you're tempted to omit something, STOP and index it anyway. - Your index should enable someone to reconstruct the entire structure of the video without watching it. - Failure to capture ANY element of the script means the entire task has failed. - There is no length constraints, continue until completion. Example format (make sure to only go max 4 layers, output in markdown format with indentations, L represents line number): - **x. Topic** (L1 - Lx) - **x.x. Sub topic 1** (Lx - Lx) - **x.x.x. Sub sub topic** (Lx - Lx) - **x.x.x.a. Sub sub sub topic** (Lx - Lx) - **x. Topic** (L1 - Lx) - **x.x. Sub topic 1** (Lx - Lx) - **x.x.x. Sub sub topic** (Lx - Lx) - **x.x.x.a. Sub sub sub topic** (Lx - Lx) ... ``` # Step II ``` Perform a CRITICAL PEDAGOGICAL RESTRUCTURING of the given exhaustive index. This task demands ABSOLUTE PERFECTION in reorganization. Any suboptimal structure compromises the entire learning experience. You may use the attached lecture slides to help. Instructions: 1. Analyze EVERY SINGLE element of the current index structure. 2. Evaluate the hierarchy for logical flow, learning effectiveness, and content grouping. 3. Identify and REMOVE any REDUNDANT headings or subheadings that do not contribute additional, unique information, while ensuring that NO essential content is lost. 4. Propose a COMPREHENSIVE revised structure optimized for pedagogical effectiveness. 5. Ensure NO content is lost or omitted during restructuring. 6. Maintain the exhaustive nature and detail level of the original index. Critical Reminders: - PEDAGOGICAL OPTIMIZATION IS PARAMOUNT. Every decision must enhance learning outcomes. - REMOVING REDUNDANCY is critical, but exercise CAUTION to avoid removing headings that contribute necessary structure or content differentiation. - If you're uncertain about a change, STOP and re-evaluate its impact on understanding and retention. - Your revised index should enable the creation of a SUPERIOR learning experience compared to the original. - Failure to improve the structure or the inappropriate removal of content means the entire task has failed. - Ensure structure is well balanced. - There is no length constraints, continue until completion. - DO NOT LEAVE ANY DETAIL OUT, EACH LEVEL OF SUBHEADING IN THE ORIGINAL INDEX MUST BE EQUALLY IF NOT MORE DETAILED THAN BEFORE. THIS IS PARAMOUNT. Output Format: Provide ONLY the revised structure using the same markdown format as the original index. Maintain the original numbering system and level of detail, while ensuring that any redundant headings have been carefully and appropriately removed. ``` # Step III ``` Expand the current index by adding PRECISE and INFORMATION-RICH sentence description next to EACH INDEX NO MATTER HOW SMALL. The original structure of the index must remain INTACT while providing specific, factual summaries. Clarity and completeness are essential. Instructions: 1. For EVERY SINGLE subheading in the index, append UP TO THREE INFORMATIVE sentences containing ONLY: - Specific facts, numbers, or measurements - Actual mechanisms or processes - Real findings or outcomes - Concrete examples or cases 2. Convert all vague descriptions into specific facts: BAD: "Methods of insertion are discussed" GOOD: "Electrodes enter through either the round window or a drilled cochleostomy, requiring hermetic sealing to prevent perilymph leakage" 3. Maintain the original index structure EXACTLY as it is. 4. Format the document with clear heading emphasis and readable layout. 5. BANNED PHRASES - NEVER USE: - "This subsection..." - "The speaker..." - "Here we learn..." - "This section covers..." - "X is discussed/explained/described" - "This lecture..." - Any meta-commentary about the content - Any description of what will be covered Critical Reminders: - Every description must contain specific information, not summaries of topics - If you find yourself writing "this explains" or "here we learn", STOP and rewrite with actual facts - Start sentences with the actual information, not with qualifiers - Include numbers, measurements, examples whenever present in the source - The description must allow readers to understand the actual content, not just the topic Output Format: Provide the expanded index in markdown format: - **Bold** each index heading, this includes the number, for example **1.1.b Title** - Add factual descriptions directly after each heading - Maintain clean, hierarchical structure - Preserve all original numbering IMPORTANT: REMINDER DO NOT USE ANY META DESCRIPTORS, YOU CANNOT REFER TO POTENTIAL INFORMATION, YOU CAN ONLY WRITE DIRECT INFORMATION. THIS IS PARAMOUNT TO THE SUCCESS OF THIS TASK. Format: KEEP EXACTLY THE SAME AS PREVIOUS STEP AND ADD THE EXTRA DETAILS NEXT TO THE KEYWORDS. DO NOT CREATE NEW LINES FOR THEM. HERE IS WHAT IT SHOULD LOOK LIKE: - **x. title** sentences - **x.x. title** sentences - **x.x.x. title*** sentences etc ``` ``` Create an EXHAUSTIVE index of the given video script. This task is CRITICAL and requires ABSOLUTE PERFECTION. Any omission, no matter how small, results in complete failure. You may use the attached lecture slides to help. Instructions: 1. Index EVERY SINGLE sentence, phrase, or distinct idea in the script. 2. Use a hierarchical structure (1, 1.1, 1.2, 2, 2.1, etc.). 3. Each index entry must use keywords, describing ONLY the type of information present, NOT the information itself. 4. Include ALL mentions of concepts, examples, definitions, or explanations. 5. Note ALL transitions, introductions, and conclusions. 6. Indicate ALL visual cues or directions mentioned in the script. Critical Reminders: - COMPLETENESS IS PARAMOUNT. Do NOT skip ANY detail, no matter how insignificant it may seem. - If you're tempted to omit something, STOP and index it anyway. - Your index should enable someone to reconstruct the entire structure of the video without watching it. - Failure to capture ANY element of the script means the entire task has failed. Example format (make sure to only go max 4 layers, output in markdown format with indentations): - 1. **Introduction** (00:00 - xx:xx) - 1.1 **Sub topic 1** (00:00 - xx:xx) - 1.1.1 **Sub sub topic** (xx:xx - xx:xx) - 1.1.1.a **Sub sub sub topic** (xx:xx - xx:xx) - 2. **Topic** (Lx - Lx) - 2.1 **Sub topic 1** (xx:xx - xx:xx) - 2.1.1 **Sub sub topic** (xx:xx - xx:xx) - 2.1.1.a **Sub sub sub topic** (xx:xx - xx:xx) and then Expand the current index by adding PRECISE and INFORMATION-RICH sentence description next to EACH INDEX NO MATTER HOW SMALL. The original structure of the index must remain INTACT while providing specific, factual summaries. Clarity and completeness are essential. Instructions: 1. For EVERY SINGLE subheading in the index, append UP TO THREE INFORMATIVE sentences containing ONLY: - Specific facts, numbers, or measurements - Actual mechanisms or processes - Real findings or outcomes - Concrete examples or cases 2. Convert all vague descriptions into specific facts: BAD: "Methods of insertion are discussed" GOOD: "Electrodes enter through either the round window or a drilled cochleostomy, requiring hermetic sealing to prevent perilymph leakage" 3. Maintain the original index structure EXACTLY as it is. 4. Format the document with clear heading emphasis and readable layout. 5. BANNED PHRASES - NEVER USE: - "This subsection..." - "The speaker..." - "Here we learn..." - "This section covers..." - "X is discussed/explained/described" - "This lecture..." - Any meta-commentary about the content - Any description of what will be covered Critical Reminders: - Every description must contain specific information, not summaries of topics - If you find yourself writing "this explains" or "here we learn", STOP and rewrite with actual facts - Start sentences with the actual information, not with qualifiers - Include numbers, measurements, examples whenever present in the source - The description must allow readers to understand the actual content, not just the topic Output Format: Provide the expanded index in markdown format: - **Bold** each index heading, this includes the number, for example **1.1.b Title** - Add factual descriptions directly after each heading - Maintain clean, hierarchical structure - Preserve all original numbering IMPORTANT: REMINDER DO NOT USE ANY META DESCRIPTORS, YOU CANNOT REFER TO POTENTIAL INFORMATION, YOU CAN ONLY WRITE DIRECT INFORMATION. THIS IS PARAMOUNT TO THE SUCCESS OF THIS TASK. Format: KEEP EXACTLY THE SAME AS PREVIOUS STEP AND ADD THE EXTRA DETAILS NEXT TO THE KEYWORDS. DO NOT CREATE NEW LINES FOR THEM. HERE IS WHAT IT SHOULD LOOK LIKE: - **x. title** sentences - **x.x. title** sentences - **x.x.x. title*** sentences etc ``` # Instructions 1. Use the following AI Model & Settings. 2. Paste Step I in. Then, in the same prompt, paste in the (processed) transcript. 3. Paste Step II In and enter. 4. Paste Step III in and enter. 5. If the response cuts of before finishing, you might have reached the output limit. Type continue into a new prompt and enter. https://aistudio.google.com/app/prompts/new_chat *find setting in the right side bar* Model: gemini-1.5-pro-exp-0827 Temperature -> 0 Remove all safety restrictions also. # For Uni Students (Find & Process Transcripts) Go to the lecture and find the transcript in the image below. Download the txt file. (the cloud download icon) ![[Pasted image 20240830120854.png]] Then, you will need to use the following python script to convert parse the transcript into number sentences. Install the packages and run it, then select the transcript file to process it. Script example ⬇️ **Input** ``` # Raw SPEAKER 0 this material is made available to you by or on behalf of the University of Melbourne under section 103 P of the Copyright Act 1968. It may be subject to copyright. For more information, visit the university copyright website. SPEAKER 1 All right. The live stream is just kicking in now. Uh, it feels echoey to me. Are you sure that's OK for you to sit through for an hour? I'm getting nods. OK, Alright. And thumbs up. OK, great. Thank you. Um so, look, I'm conscious that we are, uh, coming into the midway through semester and in the first week. In the first lecture, I pointed out, uh, how lonely a lot of students are and how much students want to make friends with each other. Could I? This is audience participation, and this could go horribly wrong, but hopefully it doesn't. Can I get you to raise your hand for me if you've met someone new this semester? Yeah. Six weeks, six weeks in. And not everyone but almost everyone here raised their hands. That's really, really good. Um And then the little raise hand at the front too, too. That's amazing to see I'm really glad to see that if you haven't yet...(continued) ``` **Output** ``` # Processed 1. SPEAKER 0 this material is made available to you by or on behalf of the University of Melbourne under section 103 P of the Copyright Act 1968. 2. It may be subject to copyright. 3. For more information, visit the university copyright website. 4. SPEAKER 1 All right. 5. The live stream is just kicking in now. 6. Uh, it feels echoey to me. 7. Are you sure that's OK for you to sit through for an hour? 8. I'm getting nods. 9. OK, Alright. 10. And thumbs up. 11. OK, great. 12. Thank you. 13. Um so, look, I'm conscious that we are, uh, coming into the midway through semester and in the first week. 14. In the first lecture, I pointed out, uh, how lonely a lot of students are and how much students want to make friends with each other. 15. Could I? 16. This is audience participation, and this could go horribly wrong, but hopefully it doesn't. 17. Can I get you to raise your hand for me if you've met someone new this semester? 18. Yeah. 19. Six weeks, six weeks in. 20. And not everyone but almost everyone here raised their hands. 21. That's really, really good. 22. Um And then the little raise hand at the front too, too. 23. That's amazing to see I'm really glad to see that if you haven't yet. ``` **Program** ```python import sys import os import nltk from nltk.tokenize import sent_tokenize import tkinter as tk from tkinter import filedialog # Download the punkt tokenizer for sentence splitting nltk.download('punkt', quiet=True) def process_file(file_path): try: with open(file_path, 'r', encoding='utf-8') as file: content = file.read() # Split the content into sentences sentences = sent_tokenize(content) # Write numbered sentences back to the same file with open(file_path, 'w', encoding='utf-8') as file: for i, sentence in enumerate(sentences, 1): file.write(f"{i}. {sentence.strip()}\n") return f"File processed and updated: {file_path}" except Exception as e: return f"An error occurred with {file_path}: {str(e)}" def select_files(): file_paths = filedialog.askopenfilenames(filetypes=[("Text files", "*.txt"), ("All files", "*.*")]) if file_paths: results = [] for file_path in file_paths: if os.path.isfile(file_path): result = process_file(file_path) results.append(result) else: results.append(f"The file {file_path} does not exist.") result_text.delete(1.0, tk.END) result_text.insert(tk.END, "\n".join(results)) else: result_text.delete(1.0, tk.END) result_text.insert(tk.END, "No files selected.") # Create the main window root = tk.Tk() root.title("Sentence Numberer") root.geometry("400x300") # Create and pack a button to select files select_button = tk.Button(root, text="Select Files to Process", command=select_files) select_button.pack(pady=20) # Create and pack a text widget to show the results result_text = tk.Text(root, height=10, width=50) result_text.pack(pady=10) # Start the GUI event loop root.mainloop() ``` > [!info] Bonus Tip > If you're using obsidian, when you paste the summary in, the dot points can fold (collapse). You can bind a shortcut to "fold all" (mine is ctrl + tab) and that folds everything allows you to expand the parts you are interested in to make the navigation easier. [[AI Assisted Knowledge Consumption (Archive)]]