$\star$ Show how to construct a set of $n$ events that are pairwise independent but such that no subset of $k > 2$ of them is mutually independent.

Get a dice with $n^2$ sides (you can actually construct one of those). Assign the numbers $1 \ldots n$ to $n$ colors, non of which is black. Use each color to paint $n-1$ sides, paint one side striped with all colors, and paint the rest black. Here's a table:

1 | 2 | 3 | ... | n | |
---|---|---|---|---|---|

1 |
1 | 2 | 3 | ... | n |

2 |
1 | 2 | 3 | ... | n |

3 |
1 | 2 | 3 | ... | n |

... | ... | ... | ... | ... | n |

n-1 |
1 | 2 | 3 | ... | n |

n |
all | - | - | ... | - |

Counting the side painted with all colors, there are $(n-1 + 1)/n^2 = 1/n$ ways to get each color. For any two colors, there are $1/n^2$ ways, which satisfies the equation for independence. However, for $k$ colors, the chances are too $1/n^2$, where they should be $1/n^k$ if that event was independent.

Thus, each pair is independent, but each subset with three or more elements is not.

I'm actually quite proud of this solution.

There is an interesting article in the Pi Mu Epsilon Journal, vol 9, 1993