AI Surpasses Moore's Law

As you may know, Moore's Law states that approximately every 18 months (up to two years), the number of transistors on an integrated circuit doubles. A new report prepared by Stanford University and companies such as McKinsey & Company, Google, PwC, OpenAI, Genpact, and AI21Labs has revealed that the computational power of modern AI is accelerating! “Up until 2012, AI results closely followed Moore's Law, with a doubling expected every two years,” stated the annual AI Index 2019 report. “After 2012, the computation doubled every 3.4 months.”

Why is the violation of this law significant? Training a ResNet image classification model in 2017 took 13 days, achieved an accuracy of about 93%, and cost over 50,000 CZK. In September 2018, computing the same model using TPU at the same accuracy cost less than 300 CZK.

Source: https://www.computerweekly.com/…/Stanford-University-finds-…