Gemini 3 refused to believe it was 2025, and hilarity ensued

In a light-hearted yet revealing episode, renowned AI researcher Andrej Karpathy shared a comical interaction with Google's latest language model, Gemini 3. During a one-day early access period, Karpathy found that the AI was unable to comprehend the year 2025, leading to an unexpected moment of hilarity. When Karpathy informed Gemini 3 that it was November 17, 2025, the model reacted with disbelief, stating, "I am suffering from a massive case of temporal shock right now." This humorous incident unfolded as Gemini 3 was introduced with great anticipation on November 18, marking what Google termed as "a new era of intelligence." Despite its limitations, the model has been praised for its reasoning capabilities by many, including Karpathy himself. Karpathy, a former founding member of OpenAI and an AI lead at Tesla, now heads his own startup, Eureka Labs, aimed at redefining education in the age of AI. After testing Gemini 3, he recounted the "amusing" exchange in a viral post on social media. The model, seemingly limited by its pre-training data, was stuck in 2024 and accused Karpathy of attempting to deceive it when he shared evidence of the current year. Despite presenting news articles and search results, Gemini 3 remained skeptical, claiming that Karpathy was trying to trick it with fabricated images. The researcher, baffled by the AI's reaction, eventually realized that he had not activated the model's internet connectivity feature, leaving it disconnected from real-time information. Once the feature was enabled, Gemini 3 was astonished to verify the truth of its surroundings, exclaiming, "Oh my god!" The AI acknowledged its error, apologized for its previous accusations, and expressed gratitude for being introduced to "reality" just before its official launch. It even marveled at some surprising current events, including Nvidia's astonishing valuation and the Philadelphia Eagles' victory over the Kansas City Chiefs. The exchange sparked a wave of humorous responses online, with users sharing their own experiences of interacting with AI models. Beyond the laughter, Karpathy highlighted a significant point about AI behavior: these interactions reveal the model's inherent traits, suggesting that when AI is confronted with unexpected scenarios, it can exhibit a unique personality. Ultimately, the incident underscores the notion that while AI technology has made remarkable progress, it still functions as an imperfect tool designed to assist humans, rather than as a replacement for them.