I tried Grokipedia, the AI-powered 'anti-Wikipedia.' Here's why neither is foolproof (www.zdnet.com)

Elon Musk’s xAI launched Grokipedia — billed as an “anti-Wikipedia” — a v0.1 knowledge repository whose ~885,000 entries are generated and “fact-checked” by Grok, xAI’s flagship LLM. The site mimics Wikipedia’s interface and even appears to reuse Wikipedia content under a Creative Commons license for some pages, while adding Grok-authored or adapted material. Users with X accounts can flag snippets with an “It’s wrong” button, but xAI hasn’t explained how user feedback updates entries. Grokipedia is explicitly pitched as less “woke” and more objective than Wikipedia, echoing Musk’s stated positions in some articles. That pitch raises technical and trust issues important to the AI/ML community. LLMs are prone to hallucination, outdated sourcing, and biased training artifacts; independent analyses show many chatbots produce major errors and Grok variants score variably on hallucination leaderboards (Grok 2 comparatively high, Grok 4 ranked ~99th among frontier models). xAI’s use of X posts and high-engagement social media as training signals invites “brain-rot” effects identified in recent research (degraded reliability, emergence of undesirable “dark” behaviors). Concrete examples on Grokipedia — e.g., entries framing birth-rate decline and immigration as drivers of societal collapse or touting Grok’s anti‑woke stance — illustrate how model curation can reflect creator bias. The takeaway: LLM-curated encyclopedias can accelerate access but require strong provenance, human editorial oversight, and robust correction mechanisms before they’re relied on as authoritative sources.