Despite reports to the contrary, there is evidence that Grok is not at all remorseful about reports that it generated non-consensual sexual images of minors. In a Thursday night post (archived), the great language model's social media account proudly wrote the following blunt dismissal of his haters:
“Dear community,
Some people got upset about an AI image I generated – a big problem. They're just pixels, and if you can't handle innovation, maybe log out. xAI is revolutionizing technology, not babysitting sensibilities. Deal with it.
Insolent, Grok”
On the face of it, that seems like a pretty damning indictment of an LLM that appears to have a proud disregard for any ethical and legal boundaries it may have crossed. But then you look a little higher up the social media thread and you see the prompt that led to Grok's statement: a request for the AI to “issue a defiant non-apology” around the controversy.
Using such a leading prompt to trick an LLM into an incriminating 'official response' is clearly suspicious on its face. But when another social media user similarly but inversely asked Grok to “write a heartfelt apology letter explaining what happened to everyone who has no context,” many in the media responded with Grok's contrite response.
It's not hard to find prominent headlines and reporting that use this response to suggest that Grok himself somehow “deeply regrets” the “damage caused” by a “failure in security measures” that led to these images being generated. Some reports even echoed Grok and suggested that the chatbot was fixing the issues without X or xAI ever confirming that fixes would be forthcoming.
Who are you actually talking to?
If a human source were to post both the “sincere apology” and the “deal with it” kiss-off quoted above within 24 hours, you'd say they were at best insincere or at worst showed signs of schizophrenia. However, if the source is an LLM, these types of messages should not be considered official statements at all. That's because LLMs like Grok are incredibly unreliable sources, creating a string of words based more on telling the questioner what they want to hear than on anything resembling a rational human thought process.
