Anti -Semitic and racist observations, Hitler praise, conspiratory publications … In recent days, Grok, Elon Musk’s artificial intelligence has multiplied the outflows of the roads. And, while X (former Twitter) eliminated most problem messages, XAI, the company behind the chatbot, wanted to apologize.
“Don’t be afraid to surprise”
According to XAI, the numerous landslides were caused by an update, carried out on July 7. The latter would have restored an old code, today considered as “obsolete”, which helped the tool to respond to the requests of X users this week. Therefore, the programmers asked the interface that “be as frank as possible” and not “be afraid to surprise the people who are politically correct.”
Another instructions: respond “as a human” by understanding the “tone, context and language” of users “to approach as much as possible. Grok thus, in some cases, sought to” validate user orientations, including hate speeches “, instead of” responding responsible or refusing to answer questionable questions “, recognizes the XAI.
Therefore, these orders “led Grok to ignore their fundamental values in certain circumstances” and produce answers “that contain non -ethical and controversial opinions.”
In fact, during the past week, many users highlight Grok’s controversial responses. The billionaire chatbot stated that Hollywood was led by a “disproportionate” number of Jews or that people with a Jewish last name were more likely. Worse, Grok said Adolf Hitler was an “effective” response to anti-white hatred. The tool has also described as Mecahitler.
Since then, everything has returned to normal. “The update was active for 16 hours (…) we eliminate this obsolete code and rework the entire system to avoid any new abuse,” adds Xai.
This is not the first time that the AI of X has produced a hate speech. Last May, Grok supported the existence of an alleged “white genocide” in South Africa. A few days later, the AI was skeptical in front of the Holocaust. Then, XAI had attributed this error to an “unauthorized modification” by an employee of the AI application (which defines his way of responding).
On July 9, Elon Musk presented Grok 4, the latest version of his chatbot, presented as “the smartest in the world.” But 24 hours after their launch, the observers wonder. According to several experts, the tool is based regularly on the positions of its boss, Elon Musk, to respond to Internet users. A disturbing fact since more and more users use Grok to verify the comments and information of a publication.
Source: BFM TV
