How can thermal fluctuations decrease entropy?

[Rodrigo Olivera]

https://en.wikipedia.org/wiki/Future_of_an_expanding_universe "Over an infinite time there could be a spontaneous entropy decrease, by a Poincaré recurrence or through thermal fluctuations (see also fluctuation theorem)"

 

[jfizzix]

That the entropy of a closed system cannot spontaneously decrease is a statement of probability rather than about dynamics.
The laws governing how particles collide and interact are reversible in time, so all processes that can happen, can also happen in reverse.
However...
The relative probability of some processes happening in reverse are so small as to be unprecedented in the history of the universe, say, like air spontaneously rushing to one side of a room, leaving the other side in near vacuum.
That said, microscopic changes in air pressure can and do happen.
These thermodynamic fluctuations are extremely small relative to the macroscopic properties of the system.
Though the probability of a large (enough to see) fluctuation is small, the probability of a small fluctuation is large.