MAGELLAN: Metacognitive predictions of learning progress guide autotelic LLM agents in large goal spaces - View it on GitHub