Can we please stop pretending that a model with 80% accuracy on a narrow task is going to revolutionize an entire field? Let's focus on actual improvements, not incremental tweaks with flashy marketing