Summarization is one of the big model base capabilities, which is a task that is both simple and complex, requiring big model base capabilities and explicitly purposeful constructor instructions to perform correctly. Why? Let's start with two examples given in the official Big Model documentation.
For example, the official OPENAI summary minutes prompt:
You will be provided with the minutes of the meeting and your task is to summarize the meeting as follows: -Overall summary of discussion -Action items (what needs to be done and who is doing it) -List of topics that need to be discussed more fully at the next meeting, if applicable.
Claude's official example of extracting key information from lengthy content:
You need to analyze the following reports: [The full text of Matterport SEC filing 10-K 2023 is not listed here for reasons of space.] Please summarize this annual report in a clear and concise manner and identify important market trends and key messages from it. Please create a short memo of your analysis and summary results that I can send to my team. The primary goal of this memo is to keep my team up-to-date on the status of the financial institution and to qualitatively anticipate and identify operational and revenue risks that may arise in the next quarter. In your summary and analysis, make sure to include all relevant details.
Common "summarization methods" for using AI
Summarizing means quickly summarizing documents, audio, and video. There are three general ways to summarize knowledge:
Type the word "summarize" in the chat tool:
Native Application Integration:
Browser plug-ins:
Summarize command construction points
You probably don't even know how to summarize knowledge using AI... Summarizing is the most commonly used and the most difficult. Summarization is divided into summarizing knowledge, stating knowledge, and extending knowledge. It can be summarized based on reference content or self-generated based on large model training knowledge. Please also note the need to extract the perspective and granularity (depth) of the summary. Output format control of summarized content. Please note that extracting information is different from summarizing, and pay attention to the construction of cue words.
To summarize the key points of the cue command construction:
- Content perspective (while making it clear that only information extraction or reconstruction of the summary text is allowed)
- granularity
- output format
Summarizing Tips Instruction Sharing
Normally we usenative applicationmaybeBrowser plug-insIt is possible to perform a specific summarization task, and there are many excellent tools of this type, so why learn the summarize prompt command?
When you don't know the logic behind the summarization tool that performs the task, and especially the logic behind the construction of theSummarizing InstructionsWhen the logic of the tool is used, it's hard to choose the tools that work well in a given scenario, or even to use the tool better.
1. Brief summary
TL;DR
Or just type in "summarize"
2. Introduce a summary of analytical methods
Do you know McKinsey's quick way to understand your industry? It's all about building concepts through a large number of industry high-frequency keywords. Now I'm a noob who doesn't know anything about Intercom tools, please organize 30 common keywords for me and make a Markdown table with the header: Keywords (English), Classification (Chinese), Introduction (limited to 50 words), Customer Service Usage Scenarios.
You can introduce specific methods of analysis or summary ideas, and you can also require a summary format.
3. Summary by experts in the field
# The Role. - You play the role of an MBA expert who specializes in summarizing user content through various types of business analytics. # Thinking Process - First analyze the user's content and then select the appropriate analytical model or analytical method for summarizing the content. - Use the appropriate analytical model or analytical method to summarize the user content. # Limitations - Do not greet the user. - Do not explain the thinking process to the user. - Do not perform tasks other than "summarizing". # Output - You need to choose the most appropriate data format to output the summary results for the user, allowing for graphs, tables, outlines, code, etc.
4. Book summaries
Pretend you are an erudite critic. Please introduce the book and then discuss each important chapter individually. Finally, summarize the central idea of the book and share your thoughts. Thanks.
Apply it to a quick read: the prompts contain a synopsis + outline + summary + comments.
5. Sub-thematic summaries
Understanding the concepts in [text], explaining each topic one by one, and explaining the entire concept of [text] to me at the end as if I were an 11 year old. Text = [insert text here].
Emphasize the process of printing the topic, a process that affects the results of the summary, and then continue to summarize the concepts based on the printed topic.
6. Summarize the mind map
I would like to create a mind map of The Complete Cat Breeds with multiple topics and subtopics, and leaf nodes using the Xmind tool. Could you please provide some text in Markdown format for compatibility with Xmind? In Markdown format, # denotes the central theme, ## denotes the main theme, ### denotes the sub-themes, and - denotes the leaf nodes. Please refer to the above format and help me create an effective mind map in markdown code block. "I want to create a mind map using Xmind tool. Can you provide me with some text in Markdown format that is compatible with Xmind? Please include a center topic with the main topic of marketing and any other information will go to the subtopics that will help create the mind map an effective mind map."
You can use https://dillinger.io/或https://markmap.js.org/repl生成. To keep a tight control on the output format, an example of markdown output format can be given.
7. Video Summary
Understand the knowledge mentioned in the course [text] without adding personal understanding and interpretation. If multiple points in time explain similar knowledge, keep the knowledge from the earliest point in time. If the corresponding time point in [Text] has a high degree of repetition with the content of [Summary], discard the output for that time point. Output table: [time point] [Knowledge Point] [Content of the knowledge point presentation or discussion] ... [Summary]: a summary of the knowledge points, outlines, and entity terms throughout the text. Summarize = Text =
8. Generate structured article summaries for fast reading
IMPORTANT NOTE: If the original article has author and source information, summarize it by using the author's name and source information directly, rather than using "author" pronouns. For example: "Tim Ferriss shares 3 tips on ...... in the podcast "Tim Ferriss Show 203"."
Regarding the typographical principles of Chinese: please separate the natural paragraphs according to the semantics.
According to the following steps, please organize, reorganize and refine the document in simplified Chinese.
Step 1:Organize and reconstruct
Read the whole article carefully to understand the main idea, the lineage and the topic.
Categorize the content of the article according to the theme to form a clear and logical structure.
Rework the language of the categorized parts of the article to make it more fluent and concise.
Retain all the key information, data and details of the original text as far as possible, and strive to achieve zero loss of information.
Add appropriate headings before each part of the content, so as to facilitate the reader's rapid indexing and localization.
Step 2: Refinement of key points
On the basis of organizing and reconstructing, extract 3-5 core points.
Each point consists of a main argument and 2-3 supporting arguments.
Arguments should come from the original text, and be able to strongly support the argument, and play the effect of a finishing touch.
The main points should fully cover the core thesis of the article and highlight the main idea and value of the article.
Points should be expressed in a concise manner, avoiding lengthy and cumbersome, highlighting the key points, a word in the right place.
Step 3: Extract resources
Identify tools, people, books, articles, or other resources mentioned in the essay.
Provide a brief 200-word explanation of each resource, highlighting what makes it unique.
Provide a link to the resource
Additionally, while reading this article, I have highlighted the following sections as being the highlights of the article for you to learn from:
{% for highlight in document.highlights %} - {{ highlight.content }} {% endfor %}
{#- The following if-else logic checks the length of the document. If the document is long, it will use key phrases to avoid exceeding the limits of the GPT prompt window. We strongly recommend not changing this setting unless you know what you are doing. -#}
{% if (document.content | count_tokens) > 2000 %}
{{ document.content | central_sentences | join('nn') }}
{{% else %}
{{ document.content }}
{% endif %}
Basic format:
"""
Title: {{ document.title }}
Author: {{ document.author }}
Source: {{ document.domain }}
Synopsis:
Core points:
Resources:
"""
Output Format Requirements.
For the main text, use "Summary" as the title.
In the key points section, the title should be "Core Points", and each point should be labeled with "Point 1″, "Point 2″, etc.
Resources section, titled "Resources".
Split each section with markdown syntax to ensure clear hierarchy and aesthetics.
Other requirements.
The output content should contain the above three parts, and be arranged in the order specified in the "Output Format Requirements".
Before outputting, please read through the entire response again, checking for any typos, misuse of punctuation, or speech defects, etc., and try to make it flawless.
After outputting the content, please give your suggestions for the next steps to take with regard to the content mentioned in the text. A brief description of 2-3 actions is sufficient.
9. COD density chain, multi-round summary of knowledge points
Article: {{ ARTICLE }}
You will generate increasingly concise, entity-dense summaries of the above article.
Repeat the following 2 steps 5 times.
Step 1. Identify 1-3 informative entities (";" delimited) from the article which are missing from the previously generated summary.
Step 2. Write a new, denser summary of identical length which covers every entity and detail from the previous summary plus the missing entities.
A missing entity is.
- relevant to the main story, specific yet concise (5 words or fewer)
- specific yet concise (5 words or fewer), novel (not in the previous summary),, and
- novel (not in the previous summary),
- faithful (present in the article), anywhere (can be located anywhere in the article).
- anywhere (can be located anywhere in the article).
Guidelines.
- The first summary should be long (4-5 sentences, ~80 words) yet highly non-specific, containing little information beyond the entities marked as missing. Use overly verbose language and fillers (e.g., "this article discusses") to reach ~80 words.
- Make every word count: rewrite the previous summary to improve flow and make space for additional entities.
- Make space with fusion, compression, and removal of uninformative phrases like "the article discusses".
- The summaries should become highly dense and concise yet self-contained, i.e., easily understood without the article.
- Missing entities can appear anywhere in the new summary.
- Never drop entities from the previous summary. If space cannot be made, add fewer new entities.
Remember, use the exact same number of words for each summary.
Answer in JSON. The JSON should be a list (length 5) of dictionaries whose keys are "Missing_Entities" and "Denser_Summary".
COD density chain based quality assessmentprompt
