White nationalist talking points and racial pseudoscience: welcome to Grokipedia (www.theguardian.com)

Elon Musk’s xAI has launched Grokipedia, an AI‑generated encyclopedia produced and “factchecked” by its Grok large language model that now contains more than 800,000 entries. A Guardian analysis found many pages praising or sanitizing white nationalists, neo‑Nazis and Holocaust deniers—portrayals of Jared Taylor, Kevin MacDonald, David Irving, William Luther Pierce and Revilo P. Oliver are cast in sympathetic, “intellectual” terms. Entries also revive eugenic and racial‑nationalist arguments, recast slogans like the “fourteen words” in evolutionary biology language, and claim academic bias suppresses pro‑homogeneity findings. xAI’s only automated response to press queries was “Legacy Media Lies.” The story matters because it illustrates how LLMs can scale and legitimize extremist narratives and pseudo‑science at high volume while claiming internal “factchecking.” Key technical implications: autogenerated content can selectively synthesize and amplify biased source material into plausible‑sounding but misleading narratives; automatic labeling of entries as “factchecked” masks lack of transparent sourcing or human editorial oversight; and an 800k‑page corpus multiplies moderation and detection challenges. For the AI/ML community this raises urgent issues around training data curation, provenance tracking, model interpretability, content‑policy enforcement, and the societal risks of deploying models that can produce persuasive, academically styled propaganda without robust safeguards.