This team should focus on a SPIR-V back-end and remove vendor lock in from NVIDIA for CUDA IN tensor AI software. A GPL licensed AI library without GPU acceleration isn't attractive outside academia.