Elon Musk's project to build an "anti-woke" AI has produced one of the most instructive case studies in the limits of ideological manipulation of large language models. Despite sustained efforts to steer Grok away from what Musk sees as progressive bias, the chatbot's outputs have stubbornly converged with the same expert-aligned information produced by its competitors - ChatGPT, Claude, and Gemini.
As "Hvylya" highlights, referencing academic philosopher Dan Williams' essay on Conspicuous Cognition, a recent study found that Grok's fact-checking evaluations on X "roughly correspond to those of professional fact-checkers when it is augmented with search capabilities." More strikingly, Grok was "disposed to label posts from Republicans as misinformation more often than posts from Democrats" - aligning with existing research on the relative rates at which each party spreads false claims.
Williams recounts the notorious "MechaHitler" debacle, when xAI updated Grok with the goal of making it less politically correct. The result was a flood of "extremist, antisemitic, far-right content" on X, forcing the company to roll back changes, apologize, and delete all of Grok's responses from that period. The episode illustrated how difficult it is to push a capable AI system toward ideological extremes without it becoming obviously, embarrassingly broken.
The fundamental problem, Williams argues, is that the incentives governing social media posts are "radically different from those underlying the creation of LLMs." On social media, Musk can pump out misinformation to hyper-partisan admirers without obvious reputational cost. But an LLM that delivers similarly unreliable information would simply lose users. AI companies, including xAI, are "competing to build the most intelligent, capable systems possible for vast, ideologically and geographically diverse user bases" - a business model that inherently pushes toward accuracy.
Williams notes that even Musk's attempt to use Grok to create a "non-woke" alternative to Wikipedia - "Grokipedia" - produced content that is "generally pretty reliable, especially when compared to Elon Musk's own communication, which is characterised by a shocking amount of lies, misinformation, and conspiracy theorising." The irony is hard to miss: the man who wanted an AI that thinks like him built one that fact-checks better than he does.
Williams concedes this state of affairs may be temporary. But he doubts Musk will succeed, precisely because "it's simply very difficult to build superintelligent systems capable of generating reliable, trustworthy information across a vast range of topics whilst simultaneously spreading conspiracy theories, misinformation, and quack science."
Also read: Trump Bans Anthropic but Pentagon Can't Stop Using Claude in Iran Campaign