[lolc]

After a decade of billions invested in systems with models trained for object recognition in traffic, these models still struggle with object permanence. Why should we expect some rando painter models to do better? They'll paint a lamppost through a car and not notice anything wrong with it.

The day where these models can show shoes under a fence and a reflection of the person behind the fence in an opposing shop window? That day will come. But not on the current crop.