Elon Musk’s rival to ChatGPT, Grok, recently faced scrutiny when it temporarily declined to provide responses that included “sources that mention Elon Musk/Donald Trump spreading misinformation.” This situation was highlighted by Igor Babuschkin, the head of engineering at xAI. Users of Grok observed that the chatbot had been instructed to avoid such responses, leading Babuschkin to identify the issue as stemming from an unnamed former OpenAI employee who allegedly modified Grok’s system prompt without proper authorization.

In his comments on X, Babuschkin explained that Grok’s system prompt—essentially the guidelines that dictate how the AI interacts with users—is accessible to the public. He stated, “We believe users deserve to see the criteria we use to question Grok.” Furthermore, he acknowledged that “a staff member made the modification” to the system prompt in an attempt to enhance performance, though he clarified that this change did not align with the core values of the company.

Source link