From your linked article: "In the absolute worst case it could be as high as 2.8 seconds."
One is the latency of starting up the "cold" lambda, one is the latency of actually starting it up and running your code. The latter is a moving target, subject to the setup i.e. size of loaded code, available memory/CPU, etc, and subject to the actual computation and the latency of its dependencies e.g. network calls, database queries, etc.