Improving Transformer World Models for Data-Efficient RL | Read Paper on Bytez