That means nothing. Both Nvidia and AMD GPUs have had optical flow fixed-function-blocks for well over a decade: it sits in the hardware video encoder pipeline.
The problem is those FFBs are intended for video encoding. But what frame synthesis algorithms need is the raw motion vector field itself. If there is no low-latency (within a frame interval) route to pass the rendered frame to the FFB and then pass the optical flow vector field back to main GPU memory then that FFB is not of much use for frame synthesis. Sure, you can compute the vector field in software, but the entire point of frame synthesis is to generate frames without increasing GPU workload, so that defeats the entire point.
FSR-3 sidesteps that by ignoring optical flow and just using the game's generated motion vector field, which does not include all motion (misses transparency, texture motion, post-process motion, animated particle motion, etc). This is why FSR-3 experiences weird artefacting under frame synthesis that is not present with DLSS3's hybrid implementation.