Elon Musk: Grokipedia, OpenAI betrayal, and free speech

Musk explains that legacy Twitter heuristics are being removed and replaced by Grok-driven ranking. This initially exposed bugs (e.g., over-amplifying content you engage with and under-showing followed accounts), but the goal is for Grok to literally read ~100 million posts per day, deeply understand them, and match them to users. A curated Following tab powered by Grok will highlight the most interesting posts from people you follow, and X will add true semantic search across text, images, and video.

xAI trained a Grok model specifically for critical thinking—breaking arguments into axioms, checking consistency, and inferring valid conclusions. That model then iterated over roughly the million most popular Wikipedia pages, correcting errors, adding context, and expanding them using broader internet research. Musk claims Grokipedia bios are substantially longer, more accurate, and less driven by grudges, and that technical pages (e.g., physics) are already “far better” than Wikipedia, with plans to add image and AI-generated explainer video content.

Every non-ad post on X has a Grok icon; a single tap prompts Grok to analyze and cross-check the content, and users can then drill deeper. This runs in parallel with Community Notes, whose logic and data Musk notes are fully open-source, and whose notes only publish when users who typically disagree still agree on the correction. Musk argues this dual system makes X “the best source of truth on the planet,” especially as Grok gets better at piercing propaganda.

Musk recounts founding, naming, and funding OpenAI as an open-source, nonprofit counterweight to Google, with incorporation documents barring founders and officers from financial gain. He claims there is a “mountain of evidence” that OpenAI was explicitly framed as open-source and nonprofit until it became clear there were enormous profits to be made. He criticizes its current structure as “Closed for Maximum Profit AI,” says its “open-source” models are essentially nonfunctional fig leaves, and expects a California jury trial in early 2025.

He and David Sacks recount how bans, shadow bans, and government coordination with Twitter’s Trust & Safety team were uncovered via the Twitter Files. Musk says FBI agents were systematically submitting takedown requests, contradicting prior testimony that shadow banning didn’t exist. His current policy is to follow the law in each jurisdiction but not go beyond it with “ideological” censorship, acknowledging tensions in countries like Brazil where judges demanded actions that conflicted with local law.