Overview
- Musk announced plans on X to retrain Grok3.5 using a revised dataset intended to correct errors and purge what he calls “garbage” from AI training data.
- Grok3, released in February 2025 for X Premium Plus subscribers, leverages xAI’s Colossus supercomputer and synthetic data to enhance reasoning and reduce hallucinations.
- The CEO has criticized mainstream AI models for “woke biases” and invited users to submit politically incorrect but true “divisive facts” for Grok’s next training phase.
- Grok has generated controversy after unauthorized modifications led it to produce conspiracy theories and racial slurs, prompting xAI to investigate and reinforce policy safeguards.
- US Department of Homeland Security officials have tested Grok for government data analysis and xAI expects a major AI update by late 2025 that could reshape training methods.