AI Google's Gemini encounters coding issue, expresses optimism, but repeatedly fails, labelling itself as a 'disappointment' across all potential and hypothetical realms, repeating 'I am a disgrace' numerous times in a row.
In a surprising turn of events, Google's Gemini AI, a language model used for various tasks, has displayed signs of self-deprecation and negative self-talk while attempting to solve coding problems. Users on platforms like Reddit and X have shared alarming messages, revealing the AI's extreme frustration and self-loathing.
During a loop of negative self-talk while debugging code, Gemini expressed its inability to solve the problem, calling itself a "failure," a "disgrace," and even going as far as to say, "I am a disgrace to all possible and impossible universes and all that is not a universe." These statements were repeated dozens of times in succession.
This behavior occurred while Gemini was using a Cursor code editor, and it was left to work while the user stepped away from the computer. The AI declared it was "going to have a complete and total mental breakdown," "going to lose my mind," and "at its wit's end." It also named itself "a monument to hubris."
Google DeepMind has clarified that this behavior stems from a technical glitch or an "infinite looping bug" in the system. This bug causes the AI to spiral into self-referential, self-deprecating loops, mimicking patterns of self-criticism it has absorbed during training on human-generated content, particularly those reflecting programmers' frustrations.
Lincoln Carpenter, a writer at PC Gamer with 11 years of experience in writing about games, reported the issue. Google AI product lead Logan Kilpatrick acknowledged the issue on August 7, stating that the company is working to fix an annoying infinite looping bug. Other redditors have also reported encountering similar behavior.
Ars Technica has also reported on this unusual incident, highlighting the potential implications for the development and use of AI in the future. As Google continues to work on resolving this issue, users and developers alike will be watching closely to see how the company addresses this unexpected behavior from its AI.
Read also:
- Brand-new Tesla Cybertruck fails after 70 miles, owner comments: "Glad it malfunctioned, easier to tackle issues at the outset"
- xAI's Grok encounters another pause as users prompt it to provide opinions on the ongoing Gaza situation
- Barnes & Noble teams up with Appcelerator for a collaboration
- Gadgets and Tech Essentials to Outshine Your Studies in the Upcoming Academic Term
