Valiant's seminal work in 1976 showed that depth-3 circuits non-trivially capture unrestricted computation. In particular strong exponential lower bounds for depth-3 circuits imply non-linear lower bounds for (almost) general circuits, a frontier in complexity theory. However, the state-of-the-art lower bounds have not changed since the 80s. In the past few years, we have developed a new program and techniques to make progress in this front. In this talk I will outline these results.