Artificial intelligence has been built on vector arithmetic but recent advances show that binary representations such as neural hashes can outperform vectors for certain AI applications without significant accuracy trade off. Hash functions map arbitrary size data to fixed-size values and are probabilistic in nature, allowing them to trade off accuracy, data storage size, performance, retrieval speed, and more. However, traditional floating point numbers have limitations due to their binary representation, which can lead to vastly different numerical changes having virtually zero impact on model predictions. Locality sensitive hashing (LSH) techniques have been used to address this issue but newer neural hash algorithms have found additional advantages by using neural networks to create optimized hashes that balance speed and information resolution. These new approaches are being applied to various areas, including approximate nearest neighbors for dense information retrieval and search indexes, with potential to revolutionize the field of artificial intelligence.