Reinforcement learning is an area in unsupervised learning that deals with learning out of experience. The machine can be described as an agent, taking actions in an environment and getting feedback, while trying to maximize some reward. Deep learning models achieve very impressive results in reinforcement learning for playing games such as chess or AlphaGo, and recently even won some world class players. In this talk I will cover a deep learning approach to play the 2048 game - using popular techniques such as fully connected and convolutional neural networks. We will cover a simple trick, which as it turns out brings impressive results with very few trials.