I changed my mind about Machine learning frameworks

I used to be excited about machine learning frameworks. Now I have questions. The framework consolidation is real. Three years ago there were a dozen options. Now it's basically PyTorch or TensorFlow, and PyTorch is winning on research velocity. Switched from PyTorch Lightning to pure PyTorch on my last project. Felt like going from training wheels to a real bike. More control, more responsibility, more 'why is this NaN' moments. I keep wanting to try JAX. Everyone says it's faster. Every time I sit down to write actual code in it, I spend the whole afternoon fighting the type system. Not there yet. I've been wrong about stuff like this before. Trying not to get swept up. I could be wrong. Happens regularly.