Scenario 1: Clinton, Sanders, and Trump are all running. Clinton and Sanders supporters assume Trump will lose. Clinton and Sanders spend most of their time attacking each other. Since supporters of Clinton and Sanders don't bother making a second pick (since they don't want to support their opponent, and assume Trump is going to lose anyway). Clinton and Sanders each get 32%, and Trump wins with 36%.
Scenario 2: Imagine a scenario where there are lots of candidates running. Something like the GOP and Democratic primaries this year, but everyone running in the general. Like this year, there are 5 candidates running on the left and 17 running on the right. Most voters aren't going to bother ranking all 22 candidates, and only rank 1-3. This causes the left to win simply because they have fewer candidates.
The thing is, many comments seem to assume that people would rationally rank all of their choices. But people don't really act like that. Some will, some will just pick their favorite, some will rank only the people they're familiar with, etc. The current system (with the primary acting as a run-off vote) allows people to reflect on new information (these are the top two candidates and this is what they believe) before making a choice. RCV seems to just kind of hope that people think about these things ahead of time (and people usually don't).
I don't see how FPTP would be better in either example, you'd just get the same or worse outcome.