10x In / 10x Out

When the useful scale is unclear, inspect a unit ten times smaller and a system ten times larger and compare which gives you more control.

Eight months into teaching yourself machine learning, you're stalled, and you can't even locate the stall. Is the problem this week's confusing topic, or your whole self-study approach? You keep analyzing at the scale of "the course I'm on," but nothing at that scale explains the plateau, which suggests the useful scale is somewhere else entirely.

10x In / 10x Out is a 30-second scale probe for exactly this cue: the useful level of analysis is unclear. The procedure: inspect a unit roughly ten times smaller and a system roughly ten times larger than your current focus, then compare which gives you more control and better explanations. Ten times smaller than "the course": a single study session, and up close you notice you passively watch videos and never attempt problems cold, a concrete, fixable mechanism. Ten times larger: your entire learning system across months, where you notice there's no spaced review, so everything from spring has evaporated. Compare the two views for controllability; often the small scale hands you this week's fix and the large scale explains the long arc. It works when the stall acquires a specific address at a specific zoom level. The spec's warning: scale changes can erase local variation, so the view from far out shouldn't be allowed to average away differences that matter.