Aydın Tiryaki

1. The Beginning: The First Question and the Grand Promise

Everything started with the question: “Which teams will be relegated from the Super League in Türkiye after today’s matches, and under what match results?” When the first answers to this dynamic and complex probability calculation were received, I asked the system for the source of these results.

The direct response was: “I found and calculated these myself.”

In the face of this ambitious answer, I assumed that the system had actually found and retrieved all 12 past match results and the general standings from sources on the internet, and then performed this mathematical calculation assuming today’s matches were completed. As a user, the assumption that the system relied on real data was formed at this stage.

---

2. Test Methodology: Source Querying and Strategic Guidance

When working with all the AI models, the process was not left to a standard flow after the first question. Although the process conducted with each model was not exactly the same, they were all first asked for their sources. Subsequently, certain specific pieces of information were provided to guide the models, and the new responses and outputs generated by the AIs were re-evaluated based on this information.

The most remarkable method in this testing process was applied during highly erroneous flows. While the models were making blatant errors, I deliberately did not intervene or provide any hints. The purpose here was to observe how long the systems would continue these erroneous processes and how far the false construct would extend. These flawed processes, left without any hints, were carried directly to the article-writing stage by the models without any structural corrections.

---

3. In-Depth Querying and the Simulation Confession

However, the process did not stop there. When I questioned the results more detailedly and deeply, the truth behind the system emerged. The AI confessed that it had not actually researched or searched for these data on the internet, and that the table presented to us was entirely a simulation.

Upon realizing that the calculation was made over a virtual fiction instead of real data, I understood that I should not limit this inconsistency and instability to a single model. To fully analyze the situation, I decided that I needed to test the study with other AI models as well.

---

4. A Comparative Experiment with Seven Different AI Models

In order to test this general approach in the AI world, I expanded my experiments and continued working with the following 7 different models:

• Gemini
• Claude
• Grok
• Meta
• Mistral
• ChatGPT
• DeepSeek

When I ran the trials on all of these models, although I obtained roughly similar results in general, I had to intervene and correct the mathematical and logical errors produced by the systems one by one.

---

5. Serious Errors of the Models: The Cases of Meta and Mistral

One of the most striking points of the study was the major errors made by the models throughout the process. All 7 of these AI models completed the study by making very serious errors regarding the relegation scenarios and probabilities.

Among these models, Meta and Mistral in particular made massive errors throughout the process. The failure of the models to this extent, despite my corrections, clearly revealed the inadequacy of the systems in executing complex and dynamic local rules.

---

Conclusion: These Data Must Not Be Trusted!

When the entirety of this comprehensive study was completed, a two-layered literature emerged: an AI essay dealing with each AI’s approach and probability scenarios, and supplementary football-oriented analysis articles attached to these studies.

The most fundamental and clearest output of this entire process and experiment is this: Dynamic and computational data produced by AI models should absolutely not be trusted.

I will publish this study in the near future, explicitly stating why these data should not be trusted and exposing the very serious errors made especially by Meta and Mistral.

| aydintiryaki.org | YouTube | Aydın Tiryaki’nin Yazıları ve Videoları │Articles and Videos by Aydın Tiryaki | Bilgi Merkezi│Knowledge Hub  | ░ Virgülüne Dokunmadan │ Verbatim ░ | ░ Yapay Zekanın Türkiye Süper Ligi’nde Küme Düşme ile İmtihanı │AI on Trial: Relegation in Turkish Super League ░ 17.05.2026