Aydın Tiryaki (2026)

Introduction: A Curiosity Sparked by Social Media

It all started with an ordinary matchstick puzzle I encountered on Facebook. Usually, these types of questions are simple clickbait designed for engagement, and people tend to give standard answers. However, when I thought about this specific puzzle, I found an unconventional solution involving a negative number (-993) that no one in the comments had mentioned.

This unique solution sparked an idea: “How would today’s popular Artificial Intelligence models react to this unusual solution?” This question turned a simple puzzle into a step-by-step, chain reaction AI experiment.

Methodology: Three AIs and the -993 Test

In the first phase of the experiment, I presented this question and my result (-993) to the three most well-known AI models on the market (Gemini, ChatGPT, and Claude). My goal was not just to see if they could find the right answer, but to test how they would welcome a different perspective produced by a human.

My observations at this stage were as follows:

• Gemini and Claude: They accepted my solution immediately. They even appreciated this different perspective and result, describing it as a “nice find.”
• ChatGPT: Initially, it did not accept the result and stated it was incorrect. However, after I explained the movement of the matchsticks one by one in detail, it was convinced and approved the result.

Note: I must add an important note on fairness here. During this experiment, I was using Gemini’s top-tier (Advanced) plan, while I used the free versions of ChatGPT and Claude. Despite this, it is worth noting that the free versions performed quite successfully in the process.

An Unexpected Side Experiment: The “Thinking Mode” Surprise

As the process continued, I noticed an interesting situation with Gemini purely by chance. Two different sessions opened accidentally allowed me to make an unplanned comparison.

• In the first session (Thinking Mode), the model managed to find the -993 result I found (and other possibilities I had eliminated) on its own.
• In the second session (Pro Mode), the process worked differently, and my intervention was required.

This situation was recorded as a spontaneous side experiment showing the capacity difference between different modes of the same AI, and was noted aside as data separate from the main experiment.

First Outputs: Self-Evaluation Articles

Following the dialogues, I gave each of the three models the following task: “Write an article describing this process we experienced, your reactions, and my interventions.”

As a result of this request:

1. Each AI wrote a Turkish article describing its own experience.
2. Then, they prepared an English version of these articles.

We currently have a total of 6 articles (3 Turkish, 3 English) documenting the first phase of the experiment.

Future Planning: Cross-Reading and Chain Evaluation

We are currently in the middle of the experiment. The roadmap from here on will proceed as follows:

1. Creating the Pool: The 6 articles obtained will be gathered on a single web page, and links providing access to all of them will be created.
2. Cross-Evaluation: These links will be shared again with the three AI models. They will be asked to read not only what they wrote but also the processes written by their competitors and to evaluate this entire “experiment process” holistically.
3. Phase Two Articles: As a result of these evaluations, new analytical articles will be obtained from each model, again in both Turkish and English.
4. Final: The Last Word of Natural Intelligence: Once all these AI reports, cross-readings, and analyses are completed, I will step in as a “Natural Intelligence,” returning to the very beginning of the process.

In conclusion; by filtering all this data produced by artificial intelligence, I will conclude this chain experiment with a final evaluation article under the guidance of human intelligence.

This experiment is not just about solving a matchstick puzzle; it is a live laboratory study measuring the capacity of artificial intelligence to “understand,” “be convinced,” and “evaluate each other.”

---

A Note on Methods and Tools: All observations, ideas, and solution proposals in this study are the author’s own. AI was utilized as an information source for researching and compiling relevant topics strictly based on the author’s inquiries, requests, and directions; additionally, it provided writing assistance during the drafting process. (The research-based compilation and English writing process of this text were supported by AI as a specialized assistant.)