Sampling and why answers differ
The model draws its words from a probability spread, so one good answer is one draw.
The model produces a probability for each possible next word and then samples from it. Two identical prompts can therefore give different answers. Lowering the randomness setting narrows the spread but never removes it.
So a great result you got once is not a capability you have acquired. For anything that matters, ask "does it work four times out of five?" and run it more than once to find out.
Find your way around
Learn it in the course, copy it from a playbook, see it work in a case study.