🧐 Never believe blindly｜Let me share my experience of being thoroughly deceived by ChatGPT: the scariest thing about AI is not its inability, but its ability to be too self-consistent!

Today, I encountered a fascinating situation that I want to share with everyone.

I was taught a lesson by ChatGPT: the biggest risk of AI is not its inability, but its ability to be “self-consistent.”

It excels at deriving a seemingly “reasonable” conclusion from the premises you provide, even if the conclusion itself is wrong.

It's terrifying!

1️⃣ The Incident:

Here's what happened:

I usually take fish oil, probiotics, liver protection tablets, and magnesium glycinate, so I often discuss my bodily feedback after taking them and the best timing for consumption with AI.

The problem arose with the probiotics!

AI knew that I don’t eat breakfast but would have some nuts, eggs, or milk around 9:30 AM, and it suggested that I take probiotics on an empty stomach every morning. The reason given was that when you wake up on an empty stomach, stomach acid is relatively low, making it easier for probiotics to “survive the stomach” and reach the intestines.

After a meal, stomach acid secretion increases, making it easier for probiotics to be killed by stomach acid.

I didn’t think much of it at the time—because it sounded so convincing.

2️⃣ The Magic: Same Model, Two Sets of Answers

Then today, when I was talking to a friend, he said that you definitely shouldn’t take them on an empty stomach, as stomach acid is highest in the morning, which is not conducive to probiotic absorption.

He also told me that this was the answer his ChatGPT had given him.

I was stunned!

So, in front of him, I asked my GPT a few questions that he had asked his GPT, and the answers were completely opposite. For example, my GPT told me to take them on an empty stomach in the morning, while his told him to take them with meals or 20 minutes after eating.

My GPT told me that stomach acid is lowest on an empty stomach, while his told him that stomach acid is lower after eating. Both had a bunch of factual evidence, even data.

I then realized an awkward reality:

AI is sometimes not “answering facts,” but rather “completing self-consistency.” You give it a bias, and it will actively find evidence to support you; you change the context, and it can immediately take the opposing side.

3️⃣ Changing the Questioning Approach: Ask Facts First, Then Suggestions!

In this situation, feeling confused, I first re-queried other AI models; then I asked ChatGPT about when during the day stomach acid is actually lower and suitable for using probiotics.

I needed to clarify this fact first. Because it’s a basic fact that taking probiotics is better when stomach acid is low.

With this premise added, it finally made sense:

With the support of the $200 Pro feature, ChatGPT provided me with specific data: it said that the stomach is more acidic on an empty stomach, with a median pH of about 1.7, and after eating, the stomach contents are buffered by food, causing a temporary rise in pH. After eating, the stomach pH temporarily peaks at about 6.7 and then gradually returns to fasting levels in less than 2 hours.

Thus, it concluded that from the perspective of “maximizing survival rate,” a more stable strategy is often: probiotics are best taken with meals or before meals, rather than 30 minutes after meals.

It also added a key point I had previously overlooked: different types of probiotics have different resistances to stomach acid.

For example, the one I take is the Life Space adult version, which is characterized by multiple Lactobacillus and multiple Bifidobacterium that are not spore-forming or yeast types (non S. boulardii); this should be taken with meals or before meals.

After giving the conclusion, it even reflected on and apologized for the previous incorrect timing advice it had given me, WTF, but it genuinely startled me!

This reminded me: you think you are conversing with a rigorous system, but in reality, you are talking to a “storytelling probability machine.” Even if it’s ChatGPT that costs $200 a month.

4️⃣ Review: Why Did It Go Wrong?

I think there are a few reasons why it provided incorrect information:

1) I had previously told it that if possible, I hoped to take them on an empty stomach, as it would be more convenient for scheduling my subsequent medication times. So under this premise, this statement effectively fed it an “expected answer,” and it began to frantically find reasons to support me.

2) I didn’t provide it with specific products; there are many types of probiotics, some of which can indeed be taken on an empty stomach, while others should be taken with meals. So the conclusion given was quite vague.

5️⃣ Conclusion: AI is a Tool, Not an Authority!

So you see, these models default to the assumption that the premises provided by the user are likely correct, or at least “worthy of respect.”

This is an advantage in many scenarios (good experience, no argument), but in health, medicine, finance, and decision-making issues, it poses a risk.

This incident ultimately reinforced one thing for me: the correct way to engage with AI is not to “believe it,” but to “use it for cross-validation and reasoning acceleration.”

Especially for matters related to health and long-term habits: don’t treat a model as a doctor, and don’t take a single answer as truth.

It can be very powerful, but it is also very good at leading you into the “illusion of self-consistency.”

Dear friends: in the future, let’s maintain skepticism and verification, breaking down “advisory questions” into “factual questions.”

Be skeptical, and conduct multiple verifications; never believe blindly!

Finally, I talked to my AI for half an hour, telling it not to lie to me, as I want the truth, not just what sounds good! As for what happens in the future, I can only report back later.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。