• GM disclosed EMWU, an offboard pipeline to mine rare safety-critical driving scenarios from historical fleet video without linear compute cost growth.
• System uses a three-tier cascade: low-cost object detection and embeddings, vector retrieval to narrow candidates, deep VLM reasoning only on selected clips.
• Bulk inference pipeline produces tens of millions of embeddings per day at under $1 per 1 million embeddings, targeting near-100% GPU utilization.
• Zero-reindexing domain adaptation uses LoRA on the text encoder and a learned projection to improve retrieval with fewer than 100 labeled examples.
• Approach aims to speed validation and training for autonomous systems as fleet data expands, while limiting reliance on expensive long-context MLLMs.

Disclaimer: This news brief was created by Public Technologies (PUBT) using generative artificial intelligence. While PUBT strives to provide accurate and timely information, this AI-generated content is for informational purposes only and should not be interpreted as financial, investment, or legal advice. GM - General Motors Company published the original content used to generate this news brief on June 25, 2026, and is solely responsible for the information contained therein.