Distinguish between inference scaling and "larger tasks use more compute"

A viral analysis distinguishes between two reasons AI inference costs are skyrocketing. First, models are tackling larger, more complex tasks that naturally require more compute—similar to humans taking longer for bigger jobs. Second, and more concerning, models may be using disproportionately more compute for the same-sized tasks, squeezing efficiency. This distinction is critical because only the second scenario represents true "inference scaling," which could make advanced AI economically unsustainable if costs outpace utility gains.