I think unlike Uber, the future floor for costs is probably very low. You can get insane improvements on inference costs/speed when you dedicate time and money

I think unlike Uber, the future floor for costs is probably very low. You can get insane improvements on inference costs/speed when you dedicate time and money towards optimization.