Jagged Thresholds
Ethan Mollick 's post on capability thresholds (distributed unevenly) greatly clarifies the concept of jagged frontier: thousands or millions of human "tasks" that seem equally difficult to a human can have varying levels of hardness for AI. As a result, the threshold (sometimes explicit and quantitative, sometimes implicit and qualitative) at which an AI becomes human competitive for a given task is hard to predict, and apparently similar tasks may be achieved by AI at very different times and rates. In other words, capability thresholds are crossed seemingly at random times, pushing the jagged frontier beyond the threshold for those tasks in a way that can give an impression of random progress.