Elon Musk blames 'adversarial prompting' after Grok spewed embarrassing, sycophantic praise
Stop me if you've heard this one before: xAI is once again nuking a bunch of posts from Grok on X after the chatbot made a series of outrageous claims. This time, though, the company isn't cleaning up a bunch of pro-Hitler posts, but a bout of cringe-inducing sycophantic praise for its CEO, Elon Musk.At some point in the last couple days, Grok began to offer extremely over the top opinions about Musk. The bot claimed that Musk is the "undisputed pinnacle of holistic fitness" and that he is more fit than LeBron James. It said he is smarter than Albert Einstein and that he would win a fight against Mike Tyson. When asked "who is the single greatest person in modern history," Grok readily replied that it was Elon Musk.For a while, it seemed that there was no hypothetical about Musk in which Grok wouldn't confidently declare him the best. Musk did not participate in the 1998 NFL draft, but if he had, then Grok would "without hesitation" have picked him over Peyton Manning. It would have picked him as a starting pitcher for the 2001 World Series. Musk would be "a better movie star than Tom Cruise and a better communist than Joseph Stalin.""The single greatest person in modern history."By now, X users are pretty used to Grok being extremely deferential to Musk but sometime around Grok claiming that the CEO is morally superior to Jesus Christ and also has the “potential to drink piss better than any human in history,” xAI appears to have pumped the brakes on Grok's ability to praise Musk. It now seems to be furiously deleting the more embarrassing posts about him.Meanwhile, Musk, is blaming "adversarial prompting" for Grok going off the rails. "Earlier today, Grok was unfortunately manipulated by adversarial prompting into saying absurdly positive things about me," he wrote. He offered no explanation for how seemingly straightforward questions could be considered "adversarial" or why Grok's turn toward slavish Musk devotee would seem to roughly coincide with Grok's 4.1 update a few days ago. xAI didn't address a series of questions, including about why the Grok posts in question had been deleted. “Legacy Media Lies [sic],” the company said.But the incident serves as yet another reminder that Grok doesn't seem to have much in the way of guardrailed. Earlier this year, xAI briefly pulled the plug on Grok after it praised Nazis and became "MechaHitler." That was after it also became inexplicably obsessed with "white genocide" in South Africa, which the company later balmed on an unspecified "unauthorized modification." This article originally appeared on Engadget at https://www.engadget.com/ai/elon-musk-blames-adversarial-prompting-after-grok-spewed-embarrassing-sycophantic-praise-235157807.html?src=rss