Birthday Problem Explained - Why 23 People Is Already a Coin Flip

The setup

Twenty-three people are in a room. Dev bets fifty pounds that at least two of them share a birthday. Only 6% of the calendar is covered - is the bet as bad as it looks?

Short answer: Take the other side and you lose slightly more often than you win: with 23 people the chance of a shared birthday is about 50.7%.

Compute the complement

P(at least one shared) is hard directly; P(all 23 birthdays distinct) is easy. The first person takes any day, the second avoids 1 day (364/365), the third avoids 2 (363/365), down to the 23rd avoiding 22 (343/365).

Multiply the chain and you get about 0.493. So the probability of at least one match is roughly 1 - 0.493 = 50.7%. Dev's bet is a slightly-better-than-fair coin flip, which is exactly why he is smiling.

Why intuition misses by so much

The 6%-of-the-calendar instinct compares 23 people to 365 days. But matches happen between PAIRS, and 23 people contain C(23,2) = 253 pairs, each a 1-in-365 shot. 253 nearly-independent chances at 1/365 lands close to a half - the count that matters is quadratic in people, not linear.

The same effect is why hash collisions, shared PINs and duplicate lottery picks show up far earlier than naive intuition expects: birthday-style collision counts grow with the square of the sample.

Why interviews ask it

Interviewers are checking two habits: reaching for the complement instead of a direct union bound, and recognising that pair-counts scale quadratically. Both habits transfer straight to collision and coincidence estimates on real data.