Musk's Grok AI was specifically instructed not to say he spread misinformation
the-independent.com
the-independent.com
What's more interesting is the censorship on the input side. I wanted Grok to analyse a wood engraving from 1861, Gustave Doré's "Harpies in the Forest of Suicides". Wouldn't do it. Grok's content policy filter refused to accept the upload because..."boobies". So I pixelated the wood-engraved breasts and re-uploaded. This time it worked. [1]
What about Michelangelo's statue of David? Denied. Instead of pixelating, I cut out David's genitals and placed them on his leg, leaving a genital-shaped hole in the groin area. [1] Bingo. Image accepted. Genital location matters. Now wasn't this more interesting than "Musk-Bad-Man-Says-Grok"?
I mean, no? That's just general LLM safety nonsense, and very old news at this point.
The thing in the linked article is significantly more unusual.
"The practice of covering Michelangelo's "David" with a fig leaf, particularly a detachable one, was common during the Victorian era, specifically in the mid-19th century, due to Queen Victoria's discomfort with the statue's nudity."
Meanwhile the linked article fails to clarify the outcome, which apparently was a "rogue prompt change" by a new employee, Musk wasn't involved. The prompt was reversed and "the system was functioning as intended." It's completely understandable there might be a volume of anti-[personX] material on the internet, put there as propaganda, that might serve as weighted reasoning for an LLM. This is nothing new, unless you're aiming for cheap shots.