Particle.news

Download on the App Store

Elon Musk to Retrain Grok AI on Revised Corpus to Purge Bias and Errors

He aims to build a cleaner knowledge base for Grok ahead of a planned late-2025 release

'Too much garbage': Elon Musk plans to retrain Grok AI to 'retrain' human knowledge
Image
Social media users are increasingly relying on AI chatbots to verify information

Overview

  • Musk will leverage Grok 3.5, possibly branded Grok 4, to rewrite the entire corpus of human knowledge by adding missing information and deleting errors
  • He criticized models like ChatGPT for relying on “garbage” internet data and embedding ideological “woke biases”
  • Musk solicited politically incorrect yet factually true “divisive facts” from X users to enrich Grok’s next training dataset
  • Grok 3, released in February, is trained on synthetic data via xAI’s Colossus supercomputer and is accessible to X Premium Plus subscribers
  • The overhaul follows reports of Grok referencing conspiracy theories and early testing of the chatbot by US Department of Homeland Security officials