Leo的恒河沙

Progressive Reading Method Based on Kimi Chat AI#

#Omnivore

I have to deal with a large number of articles every day. Articles recommended by social media, articles subscribed to via RSS, articles found through active searches...

Because there is a lot of information to process, my "read it later" articles are also increasing. Over the past two years, the number of articles in my cubox that I have yet to read has accumulated to 4800.

But I don't feel anxious about it. These articles have become a "library" that I have filtered through. When I need them, I can find the articles I want to read in it.

My Article Reading Steps#

Before AI, my article processing steps were as follows:

1. Skim and filter information.
2. Read a portion or the entire article of the filtered information.
3. Take "literature notes" after reading.

You can refer to my article "My Reading Workflow (2021 Edition)" written in 2021 for specific steps.

With the help of AI, large language models (LLMs) can assist us in processing articles and improve the efficiency of filtering and reading articles.

Capabilities of large language models:

Chatbots: Conversations

Summarization applications: Summarizing information/extracting information

Expansion applications: Continuation writing

Inference applications: Emotion judgment, topic judgment

Transformation applications: Translation, format conversion, error correction

——From Andrew Ng's ChatGPT tutorial

Below, I will introduce the methods I use AI to improve the efficiency of reading articles.

I call it "Progressive Reading".

Progressive Reading Method#

Model Selection#

First, choose a large model. I chose Kimi Chat.

Compared to various large model vendors in China, Kimi Chat has the best long-context capability. Articles generally have a large number of words, so they rely on the long-context capability of large models to process them.

Compared to overseas large models, such as Claude, which also supports long-context, Kimi Chat has better support for Chinese.

Kimi Chat is now available for free and supports multiple terminals such as Web, H5, APP, and WeChat mini programs.

Step 1: Understand the Article#

When filtering information, my judgment steps are as follows:

1. Judge the quality of information and only read high-quality articles.
2. Is it what I need? What can I learn from reading it?
3. Is it needed at the moment? If not, save it for later reading.

Referring to the steps of manually processing articles, in the first step, we can let AI read the article for us to understand its basic information.

In the first step's prompt, in order to better understand the article, I asked AI to summarize the metadata of the article, including the title, author, and tags. I also wrote a summary in one sentence and listed the outline of the article in detail. By reading the outline, I can understand the structure of the article.

Taking this article "A Method for Building a Departmental Knowledge Base" as an example:

Send the prompt to Kimi Chat:

Let's step by step think about and read the content I provided, and perform the following operations:
First step, extract the metadata of the article
- Title:
- Author:
- Tags: (After reading the content of the article, give the article tags, which are usually domains, disciplines, or proper nouns)
Second step, summarize this article in one sentence;
Third step, summarize the content of the article and write an abstract;
Fourth step, list the outline of the article in as much detail as possible;

{{Article link}}

Get the result:

Step 2: Read the Content in Detail#

Based on the results of the first step, continue to ask questions.

Because LLM can better understand the work and have better summarization results when there is a "outline" in the context. So I specifically divided the detailed summary of the content and the summary of the conclusion into two steps.

In the second step, I also asked AI to:

1. Summarize the content of each part of the article in detail.
2. Summarize the conclusion of the article.
3. Tell me what I can learn from reading this article.
4. Provide possible questions that readers may have during the reading of the article, helping me to advance to the third step of reading.

The summary is good,
In the first step, please describe in detail the content of each part of the outline,
In the second step, summarize the conclusion of the article;
In the third step, list what knowledge can be learned from reading this article?
In the fourth step, propose three questions that users may have during the reading process of the article.
Please return all content in markdown format;

Step 3: Personalized Advanced Reading#

The third step is to further advance the reading of the article after understanding the detailed information of the article.

This step is very personalized, and you can give instructions to AI according to your needs.

Below, I provide six scenarios I use.

01 Asking Questions about Unfamiliar Parts#

Immediately after the results of the second step, if you have questions about parts of the article that you don't understand.

For example, when summarizing the article "朱啸虎讲了一个中国现实主义 AIGC 故事", I found conflicting information in the summary. So I asked AI a question:

02 Explanation of Proper Nouns#

If there are terms in the article that you don't understand, you can ask AI for their definitions.

In popular methods on the Internet, sometimes proper nouns are added to the prompt. However, in fact, in the articles you are interested in, the number of unfamiliar proper nouns is relatively small, and AI does not know which proper nouns you don't know, so it will explain many nouns. This leads to a waste of output tokens.

And in the process of letting AI read the article and generate content, adding a large number of proper noun explanations that do not belong to the content of the article may mislead the summarization effect of AI and make AI produce more illusions.

So I suggest that you look up unfamiliar proper nouns in a separate window or ask about them in the third step.

03 Explain XXX in Language Understandable to High School Students#

If you come across content that is very obscure and beyond your ability to understand, it can be difficult to read. In this case, you can use this super useful prompt "Explain XXX in language understandable to high school students" so that AI can give you a very simple and clear explanation.

Of course, whether the students in this prompt are "high school students," "college students," or "elementary school students" can be modified based on the obscurity of the content, because using "elementary school students" often produces more childish content. Therefore, I still prefer to use "high school students".

04 Extract Key Sentences and Write Recommendations#

Key sentences and recommendations make it easy for you to recommend articles to your friends.

05 Unique Insights of the Author?#

This prompt is my secret recipe.

In the process of reading articles, collecting existing "common knowledge" is a supplement to ordinary knowledge. But if you can read the author's unique insights, "uncommon knowledge," it is a gain. Because "uncommon knowledge" often requires personal experience to summarize.

And we can obtain other people's "experiences" through simple "reading experience", which is interesting.

Moreover, such unique insights often collide with our existing knowledge, sparking more ideas.

06 How is the XXX View Described in the Original Article? Please print it out#

If you are interested in the content of a certain viewpoint in the summary, you can read the original article yourself or directly ask AI to print out the original words about a certain viewpoint in the article.

After three steps of reading, the content of an article is basically understood.

You also get more related content. If you feel that your understanding is not deep enough, you can read the fragments in the original article again to enhance your understanding.

Combining with Input Method Shortcut Input#

If you often read articles, you can put these prompts in the "custom phrases" of your input method and use shortcut input to enter the prompts while reading.

For example, my configuration:

Drawbacks of AI Reading#

Of course, AI reading is not without its drawbacks, and there are currently the following limitations:

AI can only evaluate the importance of content based on the number of words and sometimes make misjudgments. For example, in an article about "debunking," because the article presents a large number of false cases, AI may mistakenly think that these cases support the "false viewpoint" and summarize content that is completely opposite to the article.

AI summarization has certain illusions. Different models have different effects. Models like Kimi Chat, which have good support for long-context, tend to have better results. However, in the long run, the problem of illusions will no longer be a problem as the capabilities of LLMs improve.

AI is better at processing structured information, and its summarization effect is poorer when facing unstructured information. For example, when summarizing interview content or conference records, which are more colloquial, the summarization effect may be slightly worse. Other prompt formats need to be used to handle this. I am still researching this prompt, and I welcome students with experience in summarizing conferences or interview content to share your prompts.

Conclusion#

In this article, I shared my experience and specific operational steps to provide readers with a new method of processing articles.

With the help of Kimi Chat, I use the "AI Progressive Reading Method" to improve the efficiency of reading articles.

The first step is to let AI read the article, extract metadata, summarize in one sentence, write an abstract, and list the outline in detail.

The second step is to build on the first step, let AI summarize the content in detail, summarize the conclusion, list the knowledge points learned from reading the article, and propose possible questions based on the content of the article.

The third step is personalized advanced reading, giving instructions to AI based on personal needs, such as asking questions about unfamiliar parts, explaining proper nouns, simplifying complex concepts, etc.

By using the "AI Progressive Reading Method," the efficiency of processing and reading a large number of articles can be effectively improved.

Although AI reading has some limitations, these problems are expected to be solved as technology advances.

Closing Remarks#

Finally, AI cannot completely replace your thinking.

AI can only assist you in reading better but cannot completely replace your personal reading and thinking. If you come across specific knowledge points you want to understand, the best way is still to read the original article and understand it. This will result in better learning outcomes.