Indeed, as others have said this isn’t a hard requirement. Anyone with a handheld (e.g. Steam Deck) playing off a uSD card uses a device that’s an order of magnitude slower for sequential I/O

I’ve always referred to it as “productive procrastination”. Better than doomscrolling I guess …

Hey, I fixed a tap today!

I feel seen…

Thank you for this contribution! I was familiar with the idea of ML models capturing a compressed snapshot of the data, but that work on exploring its limits in DTs looks very interesting.