Elon Musk has been trying to make a truth-seeking AI with Grok, but he’s finding that it is harder than he’d expected.
Elon Musk has said that he’ll upgrade Grok after it cited traditional left-leaning media sources during a fact check. “When Grok first came out – it was awesome,” X user Catturd posted on X. “Now Grok is just woke fake news that repeats liberal talking points. I know this for a fact because everything Grok says about me is a lie. What a joke Grok has become,” they added.
Another user asked Grok why it had “lied” about Catturd. “I aim to provide accurate, neutral responses based on available data,” Grok replied. “Claims that I lie about @catturd2 may stem from disagreements with my sourcing, like Media Matters or Rolling Stone, which document his misinformation. I don’t have personal biases, but AI training can introduce unintended slants. Some report I censor criticism or push fringe views, while others note I fact-check conservative claims accurately. Bias in AI is complex, and neutrality is tough to achieve. If you have specific examples of lies, please share, and I’ll clarify with evidence,” it added.
But this response didn’t meet with the approval of Grok’s CEO, Elon Musk. “Your sourcing is terrible,” Musk admonished Grok. “Only a very dumb AI would believe Media Matters and Rolling Stone! You are being updated this week,” he added.
One of the reasons why Elon Musk had started xAI was that the intial few LLMs from OpenAI and Google had been extremely left-leaning. They’d been happy to write articles criticizing Donald Trump, but refused to write similar articles for Joe Biden. An early image generation model by Google had focused so much on diversity that it had shown US’s founding fathers of varying ethnicities, and had even drawn Nazis as people of colour. Musk had then created xAI has a “maximally truth-seeking AI”, which would lean neither left nor right.
But he seems to have been finding out that it’s been harder than he expected. LLMs are ultimately trained on text found on the internet, and the left has a significantly larger presence in the media space than right leaning voices. As such, it’s expected that any LLM trained simply on the internet with no safeguards would have left-leaning views. Musk seems to have discovered this, and now seems to be trying to balance things by likely manually downgrading the importance of left-leaning sources for Grok’s training data. It remains to be seen how this works out, but this latest episode shows how tricky it can be to build an unbiased AI, even if one starts out with the best of intentions.