The encyclopedia that was supposed to disrupt human knowledge curation hasn't changed a single entry in over three months.
The Summary
- xAI's Grokipedia, Elon Musk's AI-generated Wikipedia alternative, hasn't updated a single article since April 24th, despite claiming to be a "massive improvement" over human-curated knowledge
- Launched in October 2025 with 885,000 articles, now claims 6 million entries, but the "Recent changes" page shows zero live edits
- The silence reveals the gap between AI content generation and AI content maintenance
The Signal
Grokipedia went live in October 2025 with a burst of scale that only LLMs can deliver: 885,000 articles out of the gate. By November's v0.2 release, that number hit 6 million. The promise was simple: AI could do what Wikipedia's volunteer editors do, but faster, without the bureaucracy, without the edit wars, without the systemic bias of a self-selected group of contributors.
The problem is that knowledge isn't static. Wikipedia makes roughly 350 edits per minute. Events happen. Facts change. New research emerges. People die, companies pivot, laws pass. An encyclopedia that stops updating in April is a snapshot, not a reference work.
"An encyclopedia that stops updating in April is a snapshot, not a reference work."
What Grokipedia reveals is the difference between generation and curation. LLMs are incredible at synthesis: feed them context, get coherent output. But maintaining truth over time requires something different. It requires:
- Monitoring the world for changes to existing entries
- Evaluating which changes matter and which are noise
- Reconciling conflicting sources with editorial judgment
- Fighting vandalism, spam, and deliberate misinformation
These aren't generation problems. They're governance problems. Wikipedia solved them with humans: volunteer editors, talk pages, revision histories, and a culture of [citation needed]. Grokipedia assumed AI could replace that entire system. Three months of silence suggests otherwise.
The economics also tell a story. Wikipedia runs on $180 million a year, mostly from donations. Its marginal cost per edit is near zero because volunteers do the work. Grokipedia's marginal cost per update is compute: every refresh of 6 million articles burns tokens. If you're not seeing updates, it might be because no one wants to pay for them.
The Implication
The lesson here isn't that AI can't do knowledge work. It's that generation and maintenance are different jobs. AI is extraordinarily good at the first pass: summarize this, explain that, write 6 million articles overnight. But keeping those articles accurate requires continuous attention, which means continuous compute spend or a system that knows when to care.
Watch for the next wave of AI knowledge products to solve for updates, not just uploads. The companies that figure out selective refresh, change detection, and truth maintenance at scale will own the reference layer of Web4. Grokipedia's stasis is a roadmap of what not to build.