HyperLogLog, a probabilistic counting algorithm from a 2007 paper, estimates unique items by tracking the longest leading-zero streak in hashed IDs, requiring minimal memory compared to exact hash sets that scale linearly with data size. The method trades exactness for efficiency by using the statistical rarity of long zero streaks as evidence of scale, though a single counter produces estimates varying by 2-4x and requires multiple counters for reliability.

— via dev.to, Athreya aka Maneshwar