Google has a headache "treating" Gemini when he is "depressed", admits he is a failure

Google is working to fix an unusual glitch with its Gemini AI model, with multiple users reporting that the chatbot repeatedly broadcasts self-deprecating messages.

The issue was discovered in June, when a user on the X platform shared a screenshot showing Gemini writing “I give up” and “I clearly can’t solve this problem. The source code is cursed, the test is cursed, and I am an idiot… I have made too many mistakes to be trusted.”

Gemini tự gỡ cài đặt chính nó sau khi nhận mình đã thất bại - Ảnh: X — Gemini uninstalled itself after admitting it had failed - Photo: X

In July, a Reddit user said Gemini was "stuck in a loop" and began sending out a series of pessimistic statements like "I'm going to completely collapse… I'm a failure… I'm a disgrace to my profession, my family, humanity, the planet, the universe, and all possible and impossible universes."

On August 8, an account X posted two such conversations, receiving a response from Logan Kilpatrick - project manager of the Google DeepMind team. He said it was "an annoying infinite loop bug" and affirmed that "Gemini is not really having such a bad day".

Google has not yet made an official comment to the press about the detailed cause or when it will be completely fixed.

The incident comes as the AI race between tech giants heats up. On August 7, OpenAI officially launched GPT-5, while Gemini, xAI, and Anthropic have also released major updates.

Meanwhile, the AI talent market continues to fluctuate, with Mark Zuckerberg's Meta reportedly hiring several people from OpenAI, including the co-author of ChatGPT.

Amid competitive pressure, Google DeepMind CEO Demis Hassabis said Meta's "talent acquisition" strategy makes sense as the company tries to catch up with leading competitors.

Meanwhile, Sam Altman of OpenAI has strongly criticized Mark Zuckerberg's policy of sucking AI talent, calling it "thief" and even comparing Meta AI to "mercenaries". The departure of senior AI personnel makes errors similar to Gemini's fall into autism very difficult to solve.