most interesting thing was how the model already beats human performance without an AI. I suspect in lots of low-latency applications AI will outperform humans very rapidly just due to raw compute GPUs have