Gradient models with non-convex interactions

I will discuss some results on gradient models with non-convex interactions. In a specific example, which allows a reformulation by means of Gaussian gradient model with random coupling constants, I will show that in the absence of convexity, there could be more than one gradient Gibbs measures with zero tilt. Nonetheless, the large scale fluctuation structure in all such measures is still that of Gaussian Free Field. I will attempt to provide a sketch of some proofs. Based on joint works with R. Kotecky and H. Spohn.