Revealing Essential AI Fact-Checking Methods for AI Writing: Leveraging the Strengths of Perplexity, Copilot, and Gemini (Also as a Record for the Eve of SearchGPT)

In AI writing, fact-checking becomes crucial.
This is because if the source data is incorrect, everything will be output incorrectly.
Also, even if the source data is correct, discrepancies can occur, so we perform a final fact-check. Since this is done by AI, I refer to it as 'AI Fact-Checking' in this article. Finally, a human performs a fact-check; this is what I consider 'AI Co-creative Fact-Checking'.

In this article, I will focus specifically on 'AI Fact-Checking'.
This is essential and must-see content for AI writing, so please be sure to read until the end.

⬇️ For those who prefer watching the video, go here ⬇️

What is AI Fact-Checking?

Recently, more people are using AI to write articles. While convenient, the problem has emerged that the information AI provides is not always accurate. This is where 'AI Fact-Checking' becomes important.

Definition of AI Fact-Checking

AI Fact-Checking is the process of verifying whether the information gathered using AI tools is truly correct. Simply put, it is a process to ensure you 'do not take what AI says at face value'.

Specifically, it is mainly performed at two stages:

1. When gathering information (during research)

2. After finishing the article (during review)

During research, we use search-based AI such as Perplexity, Copilot, and Gemini. Since these AIs are linked to the latest information on the internet, they can verify the accuracy of information more efficiently than standard web searches.

During the review, we use search-based AI again to check the consistency and accuracy of the entire written article.

Why is AI Fact-Checking necessary?

There are mainly two reasons why AI fact-checking is necessary.

1. Reliability of information on the internet
   There is a vast amount of information on the internet, and not all of it is necessarily accurate. Since AI often generates answers based on this online information, if the original information is inaccurate, the AI's output will also be inaccurate.

2. AI Hallucinations
   AI has a phenomenon called 'hallucination'. This is when AI generates information that does not actually exist as if it were a fact. For example, it might create non-existent information, such as 'there was a law in medieval Europe that forbade keeping blue cats'.

For these reasons, it is important to perform fact-checking rather than using AI-generated information as is.

By performing AI fact-checking, you can create articles based on more reliable information. It also leads to preventing the spread of misinformation caused by AI hallucinations.

Currently, fact-checking based on the training data of Large Language Models (LLMs) is considered difficult. This is because the LLM itself has the potential for hallucinations. Therefore, fact-checking using search-based AI has become the mainstream.

Overview of the 3R Process

The 3R process is a method for effectively performing AI fact-checking. It is an acronym for Research, Link, and Review, which are referred to as Earth Text, Seed Text, and Flower Text, respectively.

Earth Text: The Foundation of Research

Earth Text is the stage that serves as the foundation for information gathering. Here, search-based AI is used to collect highly reliable information broadly. Since the information gathered at this stage becomes the foundation for subsequent writing, ensuring accuracy and reliability is particularly important.

Seed Text: Linking and Organizing Information

In the Seed Text stage, the information gathered in the Earth Text stage is organized and linked together. At this stage, you identify the relationships between pieces of information and build a logical structure.

Flower Text: Review and Final Check

Flower Text is the final stage for polishing the completed text. Here, you confirm the consistency and logic of the entire text while performing another fact check. At this stage, you remove misinformation and inappropriate expressions to finish the work as a high-quality final product.

Among the 3R processes, Earth Text and Flower Text are particularly important. Thorough research and accurate information gathering in the Earth Text stage form the foundation for high-quality writing. Then, careful review and final checks in the Flower Text stage produce a highly reliable final product. By performing fact-checking particularly carefully during these two stages, it becomes possible to create highly reliable text while utilizing AI.

Earth Text: Building the Foundation of Information through Research

When writing an article using AI, the most important stage is Earth Text. By building a solid foundation of information here, subsequent work becomes smoother, and you can ultimately write a highly reliable article.

Tools to Use

In the Earth Text stage, search-based AIs such as Perplexity, Copilot, and Gemini are primarily used. Because these tools are linked to the latest information on the internet, they can collect information and verify its accuracy more efficiently than regular web searches.

A tip for using these tools is to provide a prompt such as, 'Please fact-check the following text.' With this prompt, the AI will actively attempt to check the accuracy of the provided information.

In the case of Gemini, a button for re-verification via search is provided. By pressing this button, the AI automatically performs additional searches and conducts a more detailed fact check.

When using Copilot, it is important to select the strict mode. This setting allows you to obtain more accurate and reliable information.

Also, while Perplexity can be used sufficiently in the free version, switching to Pro makes it an even more powerful search-based AI. The Pro version can be used five times for free (as of July 27, 2024), so it is recommended to utilize it for checking particularly important information.

Points for Fact-Checking at the Research Stage

The most important thing in fact-checking at the Earth Text stage is to use multiple search-based AIs. By using not just one AI but multiple AIs, you can confirm the accuracy of information from more diverse perspectives.

For example, after checking certain information with Perplexity, confirm the same information with Copilot or Gemini. This allows you to correct errors that one AI might have overlooked or biases specific to a particular AI.

Furthermore, if the information provided by the AI does not match, you can compare the grounds for each answer and use it as an opportunity to search for more reliable sources.

AI Fact-Checking for Human Knowledge as Well

At the Earth Text stage, it is important to perform AI fact-checking not only on information obtained from AI but also on information based on your own knowledge and experience.

Human memory is not perfect; it can become vague over time or be remembered incorrectly. Even in your area of expertise, you may overlook the latest research results or trends.

Therefore, I recommend using AI to fact-check information even if you are certain about it. This allows you to update your knowledge with the latest information and prevent errors caused by assumptions.

By carefully performing such fact-checking at the Earth Text stage, the subsequent writing process becomes smoother, and you can ultimately write high-quality, reliable articles. Before moving to the next stage, it is important to repeat fact-checking until you are confident in the accuracy of the information you have gathered.

Yes, I will explain in detail about AI review and fact-checking for Flower Text.

Fact-Checking Flower Text with AI Review

The Flower Text stage is the final phase of the article creation process. Here, you perform a final fact-check and review of the entire completed text. Checks at this stage are particularly important and significantly influence the reliability and quality of the entire article.

Final Fact-Checking Using Gemini

I recommend using Gemini primarily for fact-checking Flower Text. The reason for choosing Gemini is its ability to process a large number of tokens (character count) at once compared to other AIs.

Since the text at the final stage is usually quite voluminous, Gemini's ability to check the entire text at once is extremely useful. By using Gemini, you can efficiently check the consistency, logical flow, and accuracy of individual facts throughout the entire text.

When using Gemini, it is effective to use a prompt such as:
Please fact-check the entire text below and point out any errors or inaccurate information.

By using such a prompt, Gemini will analyze not only the facts but also the quality of the entire text. Furthermore, you should perform an AI fact-check on the output text by pressing the re-check button. If a green mark appears, it indicates that the information exists on the internet, suggesting it is likely to be factual.

The Importance of Final Human Checks

While AI-based fact-checking is very useful, it is not perfect on its own. AI can sometimes make mistakes or overlook subtle nuances. Therefore, it is important that a human performs a final check even after the AI check is complete.

By performing both AI and human checks at the Flower Text stage, you can create higher-quality, more reliable articles.

Dealing with Hallucinations: Risk Management for Misinformation

One of the unavoidable problems when using AI to write articles is hallucination. Hallucination is a phenomenon where AI generates information that does not actually exist as if it were a fact. By appropriately dealing with this issue, you can create more reliable articles.

Leveraging Expertise to Detect Hallucinations

When writing articles on specialized topics, it is important to use your own expertise to check the AI's output. Always verify whether the information generated by the AI contradicts your knowledge and aligns with the latest research findings.

However, it is also important not to over-rely on your own expertise. You need a balanced approach where you neither trust the AI's information completely nor rely solely on your own knowledge. It is essential to assume that hallucinations can occur and perform checks with that premise in mind.

Preventing misinformation by using multiple AI fact-checking methods

You can significantly reduce the risk of misinformation by performing fact-checks using multiple search-based AIs rather than just one. For example, try checking the same information using different AI tools such as Perplexity, Copilot, and Gemini.

Comparing the results from each AI clarifies points of agreement and disagreement. If there are discrepancies in the results, it is a sign that you need to investigate further. If the results from multiple AIs match, the information can be considered highly reliable.

Commonality Extraction Method

An effective method when using multiple AI tools is the Commonality Extraction Method. In this approach, you extract the parts that are common among the information obtained from each AI and integrate the information based on those commonalities.

⬇️ The following video is a helpful reference ⬇️

It's actually a bit of an AI power-play video lol

Summary

In AI-assisted article creation, AI fact-checking is an essential and important process. By properly utilizing this method, you can significantly improve the accuracy and reliability of your information.

Performing AI fact-checks at each stage of the 3R process—Research (foundation text), Linking (seed text), and Review (flower text)—is the key to creating high-quality content. Thorough fact-checking during the research stage and final verification during the review stage are particularly important.

Hallucination is a risk that should always be kept in mind when using AI. Combining multiple strategies is effective for minimizing this risk. It is also effective to use your expertise to check AI output, use multiple AI tools to compare information, and use the Commonality Extraction Method to extract highly reliable information.

However, while AI fact-checking is a very powerful tool, the final judgment should be made by a human. It is important not to take AI results at face value, to always maintain a critical eye, and to perform additional verification as needed.

By going through such a process, you can create high-quality, reliable articles while maximizing the benefits of AI.

AI fact-checking has become an important part of information literacy in the digital age, and honing the skills to perform it properly is extremely valuable for modern writers.

By co-creating with the capabilities of both AI and humans, you will be able to provide more accurate and reliable information to your readers. By incorporating AI fact-checking into your daily writing process, you can consistently produce high-quality content that earns the trust of your readers.

PS: The illustrations for this blog are generated and captioned all at once.
Below is the power-play video lol