Yeah, I’m working at a company that traditionally makes digital signal processors for telecommunications, and even we’re using AI for actual practical applications. The current project I’m working on is applying an object detection model to detect different signal types in 2D spectrograms. The old way takes like 15 minutes to scan and detect across a wide band. This technique is likely going to be an order of magnitude faster (at least based on preliminary results) and lower-power, as you only need to capture one set of samples, then let the computer vision do most of the rest. The old way to scan for signals required taking a bunch of RF samples, which was very wasteful of time and energy, but there wasn’t really an alternative until now.
Anyhoo, all that to say I agree with you. It’s crazy to equate AI’s capabilities and potential to that of crypto.