Text this: Safe Reinforcement Learning With Model Uncertainty Estimates