[Towards Zero Failures] The Only Perspective You Need to Prevent AI Mistakes
Hello, this is Charipan 🐼💰✨
“The AI gave me wrong information, so I ended up having to look it up myself...”
“It seems convenient, but I’m afraid of mistakes, so I can’t use it for important work...”
Have you ever had an experience like that?
Actually, there is only one perspective to prevent AI from making mistakes.
Even I, having used it for 3,000 hours,
was pushed around by AI mistakes for the first 800 hours.
But since I started adopting this perspective,
mistakes have been reduced by 95%.
3 benefits you will gain from reading this article
Understand a simple perspective to reduce AI mistakes by 95%
Master how to create specific questions that bring you closer to zero failures
No longer need to spend time double-checking because you doubt AI answers
No difficult technical talk at all.
I will share a 'way of thinking' that you can use starting today.
The real reason why AI makes mistakes
First, there is something many people misunderstand.
“Because AI is smart, it will give me the correct answer.”
This is half right and half wrong.
AI certainly has vast knowledge, but it is
not good at distinguishing between facts and guesses.
It is not rare for it to return incorrect information with great confidence.
I once had an AI create a competitor analysis report,
and I almost submitted it as is.
When I checked it later, it was mixed with non-existent product names and
several incorrect numbers, which made me break into a cold sweat.
But one day, I realized.
“The one making the AI make mistakes was actually my own way of asking questions.”
The only perspective: Always be conscious of 'verifiability'
I will start with the conclusion.
The secret to not letting AI make mistakes is
to be conscious of 'verifiability'.
In other words, instead of just dumping the task by saying "tell me the correct answer,"
you should request it by saying "tell me in a verifiable format."
Let's compare with specific examples
❌ Questions prone to errors
"Tell me the benefits of this product"
→ Returns subjective content that the AI guessed on its own
⭕ Questions that prevent mistakes
"List the specifications of this product as stated on its official website in bullet points"
→ Returns verifiable facts based on official information
❌ Questions prone to errors
"Tell me the latest trends"
→ Vague and unverifiable information
⭕ Questions that prevent mistakes
"From the December 2024 industry news for XX, tell me the top 3 stories reported by major media along with their source URLs"
→ Verifiable information sources are clearly indicated
Do you see the difference?
[Case Study] 3 Cases That Changed Dramatically When Focusing on Verifiability
From here on, I will show you the "before and after" that I actually experienced.
Case 1: Creating a Market Research Report
Before: Unverifiable question
Question:
"Analyze the strategy of competitor Company A"Result:
An inaccurate report that sounded plausible but mixed guesses with facts. When I checked later, 30% was wrong
After: Verifiable question
Question:
"Extract the strategy of competitor Company A over the past year from their official press releases and IR information. Include source URLs for each item"Result:
A reliable report containing only facts. Verification work was reduced by 60%
The point that made the difference
Instead of a vague instruction like "analyze," I changed it to a verifiable instruction: "extract facts from official announcements." By not letting the AI guess and making it handle only facts, mistakes were drastically reduced.
Case 2: Technical Problem Solving
Before: Unverifiable question
Question:
"How do I fix this error in Python?"Result:
The code came back, but it didn't work. I ended up having to look it up on Stack Overflow myself anyway.
After: Verifiable Questions
Question:
List three possible causes for this error message. For each, please include a URL to the official documentation or a reliable technical site.Result:
I had the sources I needed to verify it myself and was able to reach the correct solution.
The Key Difference
Instead of asking for the answer on 'how to fix it,' by asking for 'verifiable sources,' I obtained the materials needed to make my own judgment.
Case 3: Fact-checking an article
Before: Unverifiable Question
Question:
Is the content of this article correct?Result:
The AI confidently answered 'Yes, it is correct,' but when I checked later, it was outdated information.
After: Verifiable Question
Question:
Find three reliable primary sources for the statistical data mentioned in this article. Please provide the URL for each source and quote the relevant sections.Result:
I was able to check the primary sources myself and make an accurate judgment.
The Key Difference
I changed the approach to having the AI gather materials for verification rather than relying on its 'correct/incorrect' judgment. By making the final decision myself, I was able to prevent mistakes.
3 Tips for Creating Verifiable Questions
I hope you found this helpful so far.
Now, I will share 3 tips on how to specifically create verifiable questions.
Tip 1: Specify the source
❌ 'Tell me about XX'
⭕ 'Tell me about XX using the official website (or academic papers, government statistics, etc.)'
Tip 2: Ask for citations
❌ 'Show me the data'
⭕ 'Show me the data with source URLs'
Tip 3: Separate facts from opinions
❌ "What is the evaluation of this service?"
⭕ "Regarding this service, please tell me the factual information such as objective user numbers and pricing, separately from the evaluations on review sites."
Just by being conscious of these three things, AI mistakes will be dramatically reduced.
In my experience, by being conscious of verifiability, mistakes were reduced by 95% and
verification time was also shortened by 60%.
Situations where this perspective is effective and situations to be careful
To be honest, this method also has its limits.
Situations where the effect is significant
Information gathering that requires fact-checking
Creating reports or materials
Technical problem solving
Analysis of statistical data
Situations to be careful
When urgency is high and there is no time to verify
Creative brainstorming (where ideas are more important than facts)
Specialized fields where you already have sufficient knowledge
In my experience, this perspective was effective in about 91% of tasks. Especially for content that will be "shown to others" or "used as a basis for judgment," verifiability becomes important.
[Practice starting today] What I want you to try with your next question
I have a request for you, now that you have read this far.
The next time you ask AI a question, please try adding just one phrase.
For example
"Tell me about XX"
→ "Tell me about XX with sources""Show me the data"
→ "Show me the data along with the URL""Analyze this"
→ "Extract facts from official information"
With just this, the quality of AI responses will change dramatically.
At first, you might feel it's a bit of a hassle.
But compared to the time spent re-checking after discovering mistakes later,
it is overwhelmingly more efficient.
I was skeptical at first too, but
the moment I tried it, I realized, 'Ah, this is true AI utilization.'
Summary: With the single perspective of verifiability, failures approach zero.
Finally, let's review the important points one more time.
The only perspective to prevent AI mistakes
Always be conscious of 'verifiability'
Specifically:
Specify the information source
Ask for citations
Separate facts from opinions
Just by having this perspective
Mistakes reduced by 95%
Verification time reduced by 60%
Trust in AI dramatically improved
Try keeping this perspective in mind when using AI from tomorrow on.
You will surely feel, 'My previous way of using it was such a waste...'
Please let me know the results in the comments!
I'm waiting for reports like 'Mistakes decreased!' or
'This way of using it was effective!'
🐼
Now, enjoy an AI life that approaches zero failures!
Follow me on X here▼
Follow me on Threads here ▼