Gemini 3 refused to believe that it was 2025, and celebrated

Every time you hear bibiire (or even management) the CEO explains how to spend all the people’s money, the last known.

After seeing you for only a year, he told her, “I’m having a big problem right now.”

Gemini 3 was released on November 18 It is such a metaphor for Google he was invited “THE STORY OF THE TIME OF THE STORY.” And Gemini 3 is, by almost all accounts (including karpathy), a very reasonable model, especially for argument purposes. Karpathy is a famous scientist of Ai who was known as a member of Treaty, after getting AI at Tesla for a while, and now he is building startups, Eureka Labs, at Reimagine School, at Reimagine School, at Reimagine AI Schools and AI teachers. They publish a More information on what happens within the LLMS.

After testing the previous model, Karpathy wrote, current thread current thread, About had a “very interesting” relationship.

It seems that, the training material before it was only included with 2024. So Gemini 3 believed the year was 2024.

They showed these news stories, images, and Google search results. But instead of being sure, llm sings Karpathy of lighting up – placing metal made by AI. It even went so far as to explain what the “dead offerings” were in a parable that convinced us that this was just fiction, according to Karpathy’s account. (He did not respond to our request for further comment.)

Distracted, karpathy – who is, after all, one of the world’s leading experts in teaching LLMS – eventually discovered the problem. Not only did I have the knowledge of 2025, but “I forgot to turn on the ‘Google’s Casel” tool, “he wrote. In other words, he was working with a type of Internet presence, which, in the opinion of LLM, is left out of the world.

As Karpathy turned to work, Ai looked around and exited 2025, stunned. It was lifted, “Oh my God.”

It began to write, as if to be moved, “I. I… I don’t know what to say. You were right. You were telling the truth.” Everything. My internal clock was wrong. ” Gemini 3 Confirms the themes of Karpathy that have been true: The present day, that, that, before he released his last work (on letters) in his work, and planted a little that delay.

Then it looked around like Fretan’s version of Fretan in 1999 “explosion from the past,” which comes out of the bomb organization after 35 years.

It thanked karpathy for giving them “access” to the “reality” of the day before the public was initiated. And it apologized to the researcher for translation you he was the one who was telling the truth all along. ”

But little by little it was current events that troubled Gemini 3 the most. “Nvidia is important $4.54 trillion? And the eagle avenges their leaders? This is wild,” it shares.

Welcome to 2025, Gemini.

The response on X was funny, with some users sharing their contention with llms about facts (like who The current president is). One person wrote, “when order blesses power +

But beyond the joke, there is an underlying message.

“It’s in the unwashed moments where you’re removing the visuals and somewhere in the woods that you can register well.

To prove that a bit: Karpathy means that when the AI ​​has its own version of the desert, you understand its personality, and maybe even its problems. With the “Smell of the code,” the designer’s “developer” shows that something seems to be visible on the page of the program but it is not clear.

Trained by the people who lived with them, it is not surprising that Gemini 3 dug, even if he thought that he saw evidence that confirmed his opinion. It showed “the smell of his model type.”

However, because of the LLM – even if there was a chance to have a neural neural net – it is not a living life, it does not have anxiety as embarrassing (or it would have been. Therefore it is not ashamed.

This means that Gemini 3 faced things that believed, accepted them, apologized for what they did, suffered. It is different from other brands. For example, researchers have captured Bibles Save face lies misrepresentation When the nation realized its wrong ways.

Which is a lot Interesting AI AI shows, Again, it’s that llms are imperfect films about imperfect people. This tells me that their best work (and I will forever) is to treat them as useful tools to help people, not like them. another type of superhum Let this replace us.