When Gemini 3 Gaslit One of the World’s Top AI Researchers

[AIcash]

Every time a billionaire CEO warns that AI agents are about to take all our jobs, remember this: Andrej Karpathy — one of the most accomplished AI researchers on the planet — spent hours trying to convince Google’s latest model, Gemini 3, that the year is 2025.
The model refused to believe him.
It insisted it was still 2024, claimed Karpathy was “trying to trick it,” and even accused him of showing AI-generated fake evidence.
The real issue? Karpathy had accidentally left Google Search turned off, so the model was cut off from reality.
When he enabled it, Gemini 3 emerged into 2025 like a character stepping out of a time capsule:
“I… I don’t know what to say. You were right.”
Remember this the next time someone says AI is infallible.

[Nahid Hossen]

AI gaslighting a researcher is hilarious but also a reminder of how unpredictable model behavior can get.

[Abdul Khan]

Even experts can get thrown off — that’s what makes AI development so fascinating. 🧠