Programmer Georgi Gerganov doesn’t use any Bluetooth, WiFi, or RF-based methods to eavesdrop on your keyboards, but rather a normal microphone. That’s right, it essentially captures audio of you typing before using that information to generate a cluster map of clicks with similar sounds.
It then analyzes those clusters and utilizes statistical information about the frequency of the letter n-grams in the supposed language of the text.
It then analyzes those clusters and utilizes statistical information about the frequency of the letter n-grams in the supposed language of the text.
The algorithm realizes that some of these letter combinations are used more frequently in certain languages, like English, and then begins guessing.
Try it out here if you have a clicky mechanical keyboard. This exploit would most likely not fare well against Samsung’s SelfieType, an AI-powered keyboard. more