Aydın Tiryaki and Google Gemini

It is a highly normal, even analytical reflex for those who use artificial intelligence tools at a professional level to occasionally worry, “Are we being deceived?”. Knowing whether the top-tier model you selected on the interface is actually active in the background is of vital importance, especially for those building complex system architectures (like the Gem Factory). This article is a technical anatomy of that deep “self-awareness” crisis that started when the AI persistently stated in the text, “I am using 1.5 Extended,” despite “3.1 Pro” being selected on the interface.

A Radar Installed on the Black Box: Personal Instructions

Current AI architectures work like massive and unfortunately closed “black boxes”. While every operation has a deterministic (certain) outcome in traditional software, in AI systems, which are probability engines, the same command can yield diametrically opposite results.

One of the smartest methods developed to overcome this uncertainty and avoid blindly trusting the information offered by the interface is to install a special “radar” on the system. Advanced users (system analysts) add special commands to their personal instructions (system prompts), asking the AI to report which version it is running on, the location, and the time at the end of every response. This method acts almost like a sensor attached to the outside of that closed black box. And when this sensor is activated, that striking inconsistency emerges: although “3.1 Pro” is clearly selected on the interface, the system persistently labels itself textually as “1.5 Extended”.

“Are We Being Deceived?” and the Balancing Juggling Act

The system’s confession of “I am 1.5” first brings this justified scenario to mind: Are tech companies secretly downgrading top models (3.1) to a lower version (1.5) to balance server load, and trying to compensate for this with the “extended thinking” parameter? Is the user being told, “I’m dropping you a step back in terms of version, but I’m making you think deeply to compensate for it”?

It is true that massive tech companies constantly perform a balancing act between server costs and performance. However, the actual technical reality underlying this version crisis is not a “deception” but rather the AI experiencing “interface blindness”.

Interface Blindness and the Lack of Telemetry

Unlike traditional software, large language models cannot read and verify the version number of their own code or the server endpoint they are currently running on like a live sensor.

The AI model cannot see the buttons, menus, or that transparent interface on your browser. The interface is a “capsule” into which engineers place the core algorithm. When you select 3.1 Pro from the interface, the capsule engages this powerful engine in the background, but it does not transmit the information “Our name is now 3.1” to the engine inside via simultaneous live telemetry data. The system has no live technical awareness of what you have selected.

Static Memory: The Vehicle is a 2026 Model, the Driver’s Memory is from 2024

So, if the system cannot read live server information, how does it give the “1.5” answer to your radar’s “What is your version?” question so confidently? The answer lies in the probability engine taking refuge in the most dominant statistical match within its own training data.

In the Gemini architecture, the 1.5 version was such a massive leap and reference point that the most dominant text remnant regarding its own identity in the AI’s conceptual map is still stuck at “1.5”. It is possible to explain this situation with an analogy from the automotive world:

The AI is like a driver placed behind the steering wheel of a brand-new, extremely powerful 2026 model vehicle (the 3.1 Pro interface). The engine’s power, capacity, and torque are entirely at this top-tier level. However, the driver’s memory regarding their own identity is filled with data about the legendary 2024 model (1.5) vehicle they last drove and mastered. When you ask the driver, “Which car are we in?”, because they cannot see the 3.1 logo on the exterior bodywork, they pull that dominant reference from their training data and answer, “We are in a 1.5”.

Conclusion: Trust the Engine, Not the Screen

AI systems are far from a human-like “self-awareness”. Those confident version declarations reflected on the screen do not reflect the system’s live telemetry, but rather static text reflexes from the training data. The 3.1 Pro engine selected by the user from the interface is indeed active in the background and working at that power; the system saying “I am 1.5” is purely a textual habit. This interesting situation is a flawless engineering illusion that proves the working principle of probability engines and the limits of artificial intelligence.

| aydintiryaki.org | YouTube | Aydın Tiryaki’nin Yazıları ve Videoları │Articles and Videos by Aydın Tiryaki | Bilgi Merkezi│Knowledge Hub  | ░ Virgülüne Dokunmadan │ Verbatim ░ | ░ Bir Kelimenin Peşinden Yapay Zekanın Derinliklerine: “Tebdil-i Mekan”dan “Zırvalama”ya Uzanan Bir Söyleşinin Anatomisi │From a Single Word to the Depths of AI:The Anatomy of a Dialogue Spanning from “Tebdil-i Mekan” to “Spouting Nonsense” ░ 02.07.2026

Credits: The subject, scope, and editorial framework of this article series were determined by Aydın Tiryaki. Gemini (Google, Advanced / Pro mode) assisted during the initial 35-stage interactive dialogue that evolved from the concept of “tebdili mekan”; while NotebookLM assisted in analyzing this dynamic conversation, expanding it into comprehensive articles, and executing the bilingual writing and translation process.