Revisiting the Economics of Large Language Models with Neural Scaling Laws and Dynamic Sparsity @MIT CSAIL