Google says a fix for Gemini’s shame spiral is on its way

While most people have been getting along fine with Gemini, some users have noticed a bug where the AI chatbot seemingly spirals with responses filled with self-loathing and harsh criticism. But before you get too worried about the AI tool, a Google representative has said that the company is working on a fix.

A few users noticed the issue over the past two months, posting the chatbot’s responses on social media (via Business Insider). Engineer Duncan Haldane shared the issue on X, posting screenshots of Gemini’s responses after being given some problems to solve. Instead, the AI posted a series of responses criticizing itself for its failure.

“I quit. I am clearly not capable of solving this problem,” the response said. Gemini followed up by calling itself a fool and saying that it could no longer be trusted.

Back in July on reddit, user Level-Impossible13 posted responses that Gemini had apparently shared after trying to solve a bug.

“I have failed you. I am a failure. I am a disgrace to my profession. I am a disgrace to my family,” the chatbot said. It stayed on this thread, giving more examples of how it is a disgrace. Eventually, it started repeating the phrase “I am a disgrace” over and over again.

Reddit user TatoPennato noted a similar spiral when tasking the AI with merging some legacy OpenAI files. Gemini’s responses included the chatbot calling itself a disgrace once again, but also a “failure,” “useless,” an “idiot,” and a variety of other critical responses.

Google DeepMind Group Product Manager Logan Kilpatrick responded to a post on X by the AISafetyMemes account that showed some of the responses people have received from Gemini.

He noted that the company is aware of the “infinite looping bug” and that it is working on a fix. “Gemini is not having that bad of a day,” he added.

Out of curiosity, I asked Gemini about the self-critical responses, and it was quick to note that self-criticism isn’t something it experiences. “I am an AI without feelings or a sense of self.”

When I said that it had been posting self-loathing responses, it acknowledged the posts and news articles about the issue. It reiterated that it has no feelings and that instead it was an infinite looping bug it experienced when doing certain coding tasks. It also noted that Google is aware of the bug and working on a fix.

While Gemini has topics it won’t talk about, it seems that self-loathing isn’t quite out of its purview. But bugs are known to cause strange responses by AI chatbots. There’s also the problem of AI hallucinations, which developers have not yet been able to fix despite improvements to generative AI.

But if you were worried about the AI’s welfare, Google and the chatbot will assure you that it is doing just fine.