is this not exactly what you said about the US Election tho
Sort of. It was a very harsh lesson in not conflating likelihood with certainty. So I won't say that LePen has a zero percent chance - but I will say that it is extremely, extremely unlikely she will win, and that her poll results aren't even close to the numbers Annoying Orange or Brexit was getting. Saying she'll win because Annoying Orange won is like saying you'll win the lottery because you won a close hand in blackjack. The probabilities here are most likely an order of magnitude off.
I'm pretty sure if we've learned anything from the last election, it's that polls are a piss-poor indicator of future results.
They aren't piss-poor. They generally give consistent results, and they are completely transparent about their uncertainties. Nate Silver's final prediction for the election was that Annoying Orange had a 35% chance of winning. His model was based on polls and gave Annoying Orange a higher chance than virtually every other person's model. It wasn't wrong just because the less-likely option happened.
Think of it this way: even if you have a perfectly accurate poll that accounts for every known variable, the 'output' you get from your data is a set of probabilities. If a model gives a candidate a 5% chance of winning, and that candidate later goes on to win, does that mean that your data was useless?