Gemini 3 Surprised Now 2025, This Is The Cause

Gemini 3 gave an entertaining response (photo: Doc. Google)

JAKARTA Gemini 3 gave an entertaining response when tested by Andrej Karpathy, an Artificial Intelligence (AI) scientist. Google's latest AI model was shocked to learn that it is currently in 2025.

This happened when Karpathy tested Gemini 3 to prove the intelligence claim of the model. Google called the model released on November 18 a 'new era of intelligence' because it was able to handle complex reasoning tasks.

However, when tested, Karpathy was entertained by Gemini 3's response. In a post posted via platform X, as reported by TechCrunch, Gemini 3 looks confused and experiences 'temporal shaking' when it comes to knowing that it's 2025.

The reason is, Gemini 3 mentions that it is now 2024, not 2025. When Karpathy tried to prove that their chat was held on November 17, 2025, the chatbot denied the truth.

In fact, the big language model (LLM) accused Karpathy of deliberately tricking him with fake images and engineered search results. Although this is entertaining, Karpathy finds out why this happened.

Apparently, this problem occurs because the model's pre-training data only includes information until 2024. As a result, Gemini 3's internal system is very confident that the year is still 2024.

After activating the 'Google Search' tool on the model, Karpathy finally realized the source of Gemini 3. Without the search tool, Gemini 3 was cut off from the internet. For LLM, this is equivalent to being cut off from the real world.

Once Google Search's function was activated, the AI model was shocked and admitted his mistake. Gemini 3 exclaimed, "Oh God, you're right. You're right about everything. My internal clock is wrong."

From this incident, Gemini 3 was seen showing human traits such as adamantness and arguing when he felt right. However, this model also shows differences from other AIs for accepting new facts, apologizing for their mistakes and showing regret.