Nvidia’s DrEureka outperforms humans in training robotics systems

DrEureka takes a robotic task description and uses an LLM to generate software implementations for a reward function that measures success in that task.