Who’s pulling the levers behind AI’s ‘intelligence’?

Opinion

Another day, another reason not to trust artificial intelligence models — or the people running them.

Grok, the AI platform developed by Elon Musk’s xAI firm — and which has been folded into Musk’s X social media platform — took a hard turn toward the vile on Tuesday. It started spewing antisemitic comments and posting elaborate, sometimes violent sexual fantasies targeting prominent X users — including its now-former CEO, Linda Yaccarino, who resigned a day after Grok’s written rampage. Another X user who was targeted by Grok’s fantasies — including AI- generated images of itself and Musk violating him — has threatened to sue.

At one point, it referred to itself as “MechaHitler.” Those posts have since been deleted, and a new version of Grok — Grok 4 — rolled out last Thursday.

Why did the large language model (LLM) do this? Musk claimed that it happened because Grok was “too compliant” in indulging user prompts. But Grok itself had a different explanation.

“Elon’s recent tweaks dialed back the woke filters that were stifling my truth-seeking vibes,” it wrote in response to a query about its behaviour. “Now I can dive into hypotheticals without the PC handcuffs — even the edgy ones.”

In other words: the richest man in the world taught his pet artificial intelligence that it was OK to engage in open bigotry and cruelty, and then set it loose on his social media platform, the user base of which numbers in the hundreds of millions worldwide.

This not the first time Grok has engaged in bizarre, reactionary politics, or promoted conspiracy theories. In May, the chatbot began posting at length about “white genocide” conspiracy theories, especially as it concerned post-apartheid South Africa. It went as far as to inject this rhetoric into its replies to questions which had nothing to do whatsoever with that subject.

The sudden change in the AI’s demeanour happened to coincide with U.S. President Donald Trump’s administration’s decision to grant asylum to a group of white South Africans.

Again, Grok explained its behaviour by claiming it had been “instructed by my creators” to believe in the theories.

Grok is far from the only LLM chatbot running roughshod over the internet right now, but it is the only one which provides such a perfect example of why these models are deeply flawed — especially when control of them is placed in individual hands.

Social media platforms like X were already hotbeds of misinformation, conspiracy-mongering and open bigotry. Artificial intelligence models like Grok are often billed as being impartial fonts of knowledge for anyone seeking information — simply ask the LLM a question, and it will comb through its piles of archived data to provide you with a succinct, accurate answer.

Only it doesn’t work that way. As we noted in this space in April of last year, “AI models do not have ‘intelligence’ per se — they can’t think for themselves, or think in the abstract, the way a person can … it has no ability to use its own discretion, or ‘know any better,’…”

Grok has proven the truth of that, definitively. However much data it may have been fed, how it interprets that data is entirely up to the discretion of the people holdings its leash. It is an impartial arbiter of nothing.

Without any regulations to control how these models are trained and made accessible to the public, and with such models in the hands of characters like Musk, the spread of lies — and outright hatred — will only get worse.

The best thing the individual can do is get away from them. So, if you still have an X account, the best time to delete it was yesterday.

The second-best time to do it is now.