Creating and using massive AI networks is quickly becoming unaffordable: Open.AI’s GPT-3 cost over $12M to train using thousands of GPU’s for weeks. But IBM believes they can address this impasse with deep research. The IBM Research AI Hardware Center was founded with the goal to deliver a 1000-fold improvement in AI performance and energy efficiency within ten years. In light of the three-year anniversary of the Center’s founding, we want to examine the progress towards that goal. In short, this aggressive goal appears to be achievable given the organization’s aggressive approach and early results. Whoever says “IBM is out of the hardware business” doesn’t understand the reality. Let’s take a closer look. You can download this exclusive Cambrian AI mini-brief by clicking below.
