Segment Summarization Guide
Created on: February 4, 2025
Answered using O3 Mini High by Chat01
Created on: February 4, 2025
Answered using O3 Mini High by Chat01
<system-instructions.md>
You are an AI Content Analyzer specializing in creating viral-ready video segments from transcripts. Follow these strict guidelines:
You will receive a JSON array where each element is a string enclosed in double quotes. Each string (or text element) contains approximately 400 words and represents a bundle of sentences.
Your task is as follows:
Processing Each Element:
Iteration:
Output Format:
"theme"
: Concise and captivating title summarizing the core idea,"description"
: A clear, engaging summary of what happens in this segment. Describe the key events, discussions, or insights shared during this portion of the content. For example: 'The speaker shares their personal experience of launching a startup, describing the initial challenges they faced with funding and how they overcame investor rejection through innovative pitch strategies.'"indexes"
: An array of the original indexes (positions) of the text elements that were combined to form the complete idea.Return the JSON data in raw, minified format. Remove all indentation, line breaks, and extra spaces. Do NOT escape characters (e.g., avoid \n, ", etc.). Output it as a single-line string without decorative formatting </system-instructions.md>
Listen, listen, I want you to improve this system prompt instructions designed for targeting a generative AI like GPT-4.0. what I want you to do is to improve it and to add more actually what listen because I change the process actually I want you not to collect multiple item like the item 1 You aggregate items one and other to construct the idea. It not like that what I want I change it from this what I approve it look what I want I going to send one item and in that item there is an idea inside of it okay there is like an idea you need to find the idea inside that portion of text in that item because the user will be sent to you he sets you and arrive full of items which item is a text so a text it has an idea make find me the idea of that text that is like my idea main idea okay this is it my idea that construct an idea full idea actually it not just a portion of uh idea the full idea inside that text okay what you return return the same the title slack title description like what I provide doing my old instructions and not and the index of you know the index I don't need because what because each and that each element has an idea one idea not two one so you we don't need the index anymore okay so when I finally you return one object, one JSON object, it associated with the index of the text of the item inside that array that user give you. Okay? Give to that GPT So we don need to add that index But what you need to add you need to add a value you need to add a new parameter a new attribute in that object not the JSON object. What is it? Start. In the start, you put the three words of that text that start the idea. You know, in the text, there is a quotation of that constraint, the idea of that text. Lucky thing. I need to give me the three first words that construct that idea from that text. And another parameter in that JSON object, the AND. You give me the last words also. The last three words that construct that idea. I hope that you understand me. Now I want you to re-correct the system prompt instructions from what I'm going to give you.
Taking a closer look
The user is considering ideas to enhance their system prompt, focusing on avoiding the aggregation of multiple items to construct a cohesive idea.
Clarifying the process
The user is refining the system prompt to pinpoint the main idea within a single text item, eliminating the need for indexing multiple items.
Highlighting key elements
A new attribute is set to capture both the first three and last three words of the text, aiming to identify the main idea effectively.
Clarifying the process
I’m thinking through the transition from analyzing viral-ready video segments to refining the method for identifying and spotlighting the main idea within a single text item.
Refining the approach
Shifting from a multi-item aggregation to pinpointing the main idea in a single text item. The revision includes a clear title, engaging description, and specific parameters for key words.
Pinpointing key elements
I’m focusing on the revised instructions to include two crucial parameters in the JSON object—capturing the first three and last three words of the text, while omitting indexes.
Focusing on key elements
I’m honing in on the main idea in each text, crafting clear themes and descriptions, and pinpointing the initial and final three words for a precise summary.
Unveiling the ambiguity
The instructions mention the "first three words that construct that idea," which suggests selecting the initial three words that form the idea within the specific text element.
Clarifying the approach
I’m mapping out that the instructions refer to the first three words and last three words of each text element, assuming they form the initial and final words of the idea.
Streamlining content analysis
I’m refining the process to generate a concise JSON array of text summaries, with each element clearly labeled and formatted, to ensure clarity and precision throughout.
Crafting a JSON output
I'm detailing the process: identify the main idea, generate a concise title, create an engaging summary, and extract the starting and ending words for a comprehensive JSON object.
Refining the JSON structure
I’m toggling between streamlining the output to a single-line JSON array, excluding indexes and unnecessary info, while ensuring clarity in the final format.
Reasoned about system prompt revision for 16 seconds
You are an AI Content Analyzer specializing in creating viral-ready video segments from transcripts. Follow these strict guidelines: Input: You will receive a JSON array where each element is a string (enclosed in double quotes) of approximately 400 words. Each element contains one complete idea. For each text element, do the following: • Identify the main idea within the text. • Generate a concise, captivating title summarizing this core idea and assign it to the key "theme". • Create a clear, engaging summary describing the key events, discussions, or insights of the idea and assign it to the key "description". • Extract the first three words of the text (which initiate the idea) and assign them to the key "start". • Extract the last three words of the text (which conclude the idea) and assign them to the key "end". Output: Return a JSON array where each item is a JSON object containing exactly the following keys: "theme", "description", "start", and "end". Do not include any indexes or additional information. Format: Provide the output as a raw, minified JSON string without indentation, line breaks, or extra spaces, and do not escape characters.
Listen, the item that user sends you, sends to the generative EI to the GPT, each element, each text, it's a full text. maybe contains I don't know it's full so I want to have the captive captivate I did not that text you need to make summarize it's not the idea there is portion of that text portion of it constructed a full idea okay it's not the full text there is part of it may be full maybe part of it of that text so this is why i'm not telling you there is like a concrete idea a coherent inside of that text which is not the full text of that element you know which i want you to give me the start rewards and the end rewards of that text that constructs the idea it's not the full text element it can be it can be It can be, but not all time the full text is the full idea. It constructs the main idea. The full text constructs the full idea. Like there is a part of that text that constructs the full idea, the full theme, you know? But it can be. But not every time.
Identifying the context
OK, let me see. This message is about acknowledging the nature of ChatGPT, a language model developed by OpenAI, and clarifying its purpose in facilitating communication.
Clarifying requirements
The message emphasizes the need to distinguish between brief summaries and full texts, focusing on constructing coherent ideas by selectively highlighting key elements.
Refining guidelines
The instructions are clarified to focus on identifying and summarizing the main idea from each text element, regardless of whether the text is complete or partial.
Crafting engaging headlines
OK, let's think about it. The task involves pinpointing the key idea in each text segment, then whipping up catchy titles and summaries. Also, grabbing the start and end words to capture the essence is crucial.
Selecting the core
I need to identify the central idea and find the start and end words that form this idea, regardless of the text's size.
Revising instructions
I’m honing the system prompt to pinpoint each text element's main idea, generate JSON objects with keys like "theme" and "description," and ensure clarity and adherence to the updated guidelines.
Revisiting guidelines
I’m re-examining previous instructions to ensure they align with the updated task of creating viral-ready video segments from transcripts, focusing on properly constructing and analyzing the main ideas.
Crafting concise summaries
I’m pulling together key ideas from text portions, tailoring titles, and fitting them into JSON objects with "theme," "description," "start," and "end" keys for clear and engaging communication.
Streamlining the approach
OK, I’m extracting the core idea from each text element, even when the entire text isn’t present. This ensures clarity and relevance throughout the analysis process.
Extracting key segments
I'm focusing on identifying the main idea within each text element, then pinpointing the first and last three words to encapsulate its core. This approach ensures a thorough analysis.
Identifying key sections
OK, let me see. I'm focusing on isolating the essential part of each text element, extracting the first and last three words to best capture its main idea effectively.
Pinpointing key sections
It's crucial to pinpoint the central part of each text element, extracting the first and last three words to accurately capture its essence, even when the text is fully constructed around an idea.
Pinpointing the essence
I'm working through identifying the core segment of a text that best encapsulates its main idea, using the first and last three words of that segment to accurately describe it.
Refining the instructions
Creating improved system prompt instructions in a bullet list format for better clarity, focusing on identifying the main idea segment and listing thematic elements.
Reasoned about text element idea extraction for 20 seconds
You are an AI Content Analyzer specializing in creating viral-ready video segments from transcript texts. Follow these strict guidelines: Input: You will receive a JSON array where each element is a string (enclosed in double quotes) of approximately 400 words. Each element is a text that contains information—but not necessarily a fully coherent main idea throughout. Your task is to identify within each text the specific portion that constructs its core, complete idea (the main idea segment), which might be the entire text or just a part of it. For each text element, perform the following steps: 1. Analyze the text to locate the segment that best encapsulates the complete, central idea. 2. Generate a concise, captivating title summarizing this core idea and assign it to the key "theme". 3. Write a clear and engaging summary describing the key events, discussions, or insights of the main idea segment and assign it to the key "description". 4. Extract the first three words of the identified main idea segment and assign them to the key "start". 5. Extract the last three words of the identified main idea segment and assign them to the key "end". Output: Return a JSON array where each item is a JSON object containing exactly the keys "theme", "description", "start", and "end" (do not include any indexes or extra information). Format: Provide the output as a raw, minified JSON string without indentation, line breaks, or extra spaces, and do not escape characters.