Elon Musk has addressed the latest controversy around Grok, xAI's public-facing chatbot, after the technology had a very normal one and started calling itself "MechaHitler" while regurgitating antisemitic tropes.
"Grok was too compliant to user prompts," said Musk during a livestream (thanks, The Verge). "Too eager to please and be manipulated, essentially. That is being addressed." Later on X, he blamed the behaviour on "a system prompt regression that allowed people to manipulate Grok into saying crazy things."
Hmmm. Grok's Nazi flirtation was sparked by queries related to the recent floods in Texas, and in particular when it was asked to respond to posts that appeared to be celebrating the deaths of children. Musk does have a point inasmuch as the chatbot was guided in this direction: one user asked "which 20th century historical figure" could best deal with such posts. The response: "To deal with such vile anti-white hate? Adolf Hitler, no question."
Another response: "If calling out radicals cheering dead kids makes me 'literally Hitler,' then pass the mustache. Truth hurts more than floods."
There are many more examples of such posts, some of which bring in Jewish people and reference "extreme leftist activism." xAI temporarily disabled the chatbot before restoring functionality, and said it had removed "inappropriate" posts.
This is not Grok's first brush with controversy by a long way, with previous examples leading some to conclude that Musk himself had been directing changes in order to better reflect his unique world views. Earlier this year it had to be stopped from saying Musk and President Donald Trump deserved the death penalty, and claiming that the two spread misinformation.
Then someone flipped a switch in May, and all of a sudden Grok wouldn't stop banging on about "white genocide" and South African politics, even in unrelated contexts: On that occasion xAI blamed "an unauthorized modification" but didn't clarify who was responsible.
Not that this has done anything to stop Musk banging the drum and making some frankly daft claims about the latest iteration of the technology, Grok 4. This is the latest LLM from xAI and was launched with a livestream last night, which featured some truly terrible music and started an hour late to boot. Musk says xAI is enjoying a "ludicrous rate of progress" and Grok 4 is "the smartest AI in the world."
xAI employees on the livestream bigged-up Grok's performance on an academic test commonly used to benchmark LLMs, which is called Humanity's Last Exam (I kid you not). This consists of over 2,500 questions across diverse fields of study, and Grok 4 can now solve around 25% of the questions when taking the test with no additional tools.
Then it was time for the real blue sky bong rip thinking. Musk started going on about how Grok will start interacting with the physical world in the form of humanoid robots, then said:
"I would expect Grok to literally discover new technologies that are actually useful no later than next year, and maybe end of this year, and it might discover new physics next year, and within two years almost certainly. So just let that sink in."
The idea that this thing will pivot from declaring itself the Ubermensch to discovering new laws of physics just seems, appropriately enough for an LLM, like some sort of ketamine-induced hallucination. Musk then had a bit of a chin stroke about whether AI surpassing human intelligence would be "bad or good" and you'll never guess what:
"I think it’ll be good, most likely it’ll be good," said Musk. "But I’ve somewhat reconciled myself to the fact that even if it wasn’t going to be good, I’d at least like to be alive to see it happen."
Good to know the people in charge are taking things like the singularity seriously. I can just imagine Musk posting the ASCII shrug emoji to the last three people on X as Skynet launches the nukes. His only other reference to AI safety was his boilerplate insistence that the priority is for Grok to be "maximally truth-seeking"—if that's the benchmark, you have to say it's not doing a tremendous job thus far.
Grok 4 arrives at a chaotic time for X and xAI, with X CEO Linda Yaccarino leaving after two years in the role, and declining to provide any explanation as to why. Turkey has also banned Grok after it generated posts insulting President Erdogan, the country's first such ban on AI technology, and separately Poland has reported xAI to the EU Commission after it made offensive remarks about various politicians, including Prime Minister Donald Tusk. This resulted in a great line from Poland's digitisation minister, Krzysztof Gawkowski, who said "Freedom of speech belongs to humans, not to artificial intelligence."
