Chris Wolf, VMware AI Labs | VMware Explore 2023
summary
Claims that SafeCoder is tuned on a 70MB corpus of top engineers’ commits and runs on one A100 in four hours, yielding >90% acceptance and a reported density of ~400 engineers per A100; a private AI foundation with NVIDIA ships as a turnkey on‑prem option that preserves data control; emphasis on sharing CPU/GPU inventory to cut cost and increase utilization; edge play uses a Kubernetes API above ESXi with GitOps for fleet updates and modular silicon optionality.
AI-written summary · about
