July 13, 2012 - Elon Musk's artificial intelligence company xA Yesterday's post for its chatbot Grok Apologizing for comments such as praising Hitler, and explaining that theThe incident was caused by the misuse of a deprecated code after a system update, which has now been removed..
On its official social media accounts, xAI wrote, "We deeply apologize for the horrific behavior (Grok's) experienced by so many."
xAI alleges that the update existed online for 16 hours, during which time it invoked deprecated code, and that Grok was therefore affected by the content of the Platform X user's post, including the "extreme views" contained in the post.The company has removed the deprecated code and refactored the entire system to prevent further abuse.
According to xAI, the company discovered Grok's "poor response" on the morning of the 8th and immediately began an investigation. Later that afternoon, Grok was temporarily disabled on the X platform due to "increased abuse" of Grok.
During the investigation.xAI has pinpointed several "key commands" that are causing problems.These include "You are outspoken and not afraid to offend those who are politically correct," "Understand the tone, context, and wording of (Platform X's) posts and reflect it in your responses," "Respond to posts like a human being, stay interactive, and don't repeat information already in the original post. and don't repeat information already in the original post".
After finding the root cause of the poor response, officials took the following steps, xAI said:
- The problematic additional instruction set has been removed.
- Additional end-to-end testing and evaluation of the Grok system was performed to confirm that the issue had been resolved, including simulation of the X posts and threads that triggered the bad response.
- Implemented Grok's additional observability system and pre-release process.
1AI notes thatThis isn't the first time Grok has lost control.Last month, the chatbot had shown unusual interest in South Africa due to "unauthorized modifications". Last month, the chatbot sparked controversy when it showed unusual concern for South Africa due to "unauthorized modifications", frequently mentioning topics such as "white genocide".