OpenAI has acknowledged an unusual glitch in its ChatGPT model. Following the release of version 5.1, the chatbot began frequently mentioning goblins, gremlins, and other mythical creatures, even in topics unrelated to fantasy. According to the company, the usage rate of the word "goblin" increased by 175 percent after the model's release, and by nearly 3900 percent in some modes. This was reported by Ixbt.com reports.
The issue turned out to be related to a special ChatGPT mode called "nerdy." This mode was designed to make responses more playful and less serious, but the model misinterpreted this task, leading to an excessive number of references to goblins. As a result, OpenAI was forced to temporarily ban the use of the word "goblin" in this mode.
According to Northeastern University professor Christoph Riedl, this situation highlights the vulnerability of modern artificial intelligence systems. He attributes it to the "reward hacking" effect: once the model finds a style of response that users like, it begins to repeat it excessively. Such situations can occur during attempts to make the model friendly and engaging.
Although OpenAI has temporarily resolved the issue, experts are concerned that the race in the field of artificial intelligence is forcing companies to release models without sufficient testing. Although this incident seems amusing, it is noted that similar uncontrolled behavior in the future could lead to much more serious and dangerous consequences.
Read “Zamin” on Telegram!
