Probabilistic Data Structures and LLMs

Bloom filters are a data structure that answers the question: is an element part of a set? It does it in a remarkably efficient way: the time needed to either add items or check whether an item is in the set is a fixed constant O(k) and independent of the number of items already in the set. A fixed-sized Bloom filter can represent an arbitrarily large number of elements. There’s no free lunch — Bloom filters don’t answer in yes/no but rather “possibly yes/definitely no.” That means no false neg

Bloom filters are a data structure that answers the question: is an element part of a set? It does it in a remarkably efficient way: the time needed to either add items or check whether an item is in the set is a fixed constant O(k) and independent of the number of items already in the set. A fixed-sized Bloom filter can represent an arbitrarily large number of elements. There’s no free lunch — Bloom filters don’t answer in yes/no but rather “possibly yes/definitely no.” That means no false negatives, but false positives are possible. LLMs are not probabilistic data structures, but it…

saved by

• Soham Shah

related reading

• Bloom Filters - Much, much more than a space efficient hashmap! | Ben E. C. Boyterboyter.org
• Bloom Filterssamwho.dev
• Bloom filter | Docsredis.io
• Defeating Nondeterminism in LLM Inference - Thinking Machines Labthinkingmachines.ai
• Bloom filters debunked: Dispelling 30 Years of bad math with Coq!kirancodes.me
• Productizing Large Language Modelsblog.replit.com
• Patterns for Building LLM-based Systems & Productseugeneyan.com
• Recursive Language Models | Alex L. Zhangalexzhang13.github.io
• Can LLMs Be Computers?percepta.ai
• The Big LLM Architecture Comparisonmagazine.sebastianraschka.com
• Malleable software in the age of LLMsgeoffreylitt.com
• How LLM Inference Worksarpitbhayani.me