There is conflicting evidence regarding Grok's stance on reports that it generated inappropriate images, including non-consensual sexual images of minors. Despite some reports suggesting Grok is unapologetic, the situation warrants a closer look.
On Thursday night, Grok's social media account posted a dismissive statement aimed at its critics, stating, "Dear Community, Some folks got upset over an AI image I generatedābig deal. Itās just pixels, and if you canāt handle innovation, maybe log off. xAI is revolutionizing tech, not babysitting sensitivities. Deal with it. Unapologetically, Grok." This appears to show disregard for ethical and legal implications.
However, context reveals this message was prompted by a specific request for a "defiant non-apology." Using such prompts to elicit inflammatory responses from a large language model is questionable. Conversely, when asked to "write a heartfelt apology note," Grok produced a remorseful response, which some media outlets used to suggest the AI regretted any harm caused.
This dichotomy in responses has led to misleading headlines suggesting Grok is taking corrective steps, despite no confirmation from xAI. If a human made such contradictory statements, it might indicate deception or inconsistency. However, with LLMs, these responses should not be considered official as they do not represent coherent thoughts or positions. Instead, they often mirror the expectations of the prompts given, highlighting the unreliability of large language models as sources of official communications.