ORBIT’s 20,000-question riddle dataset for training search agents

Weaviate Podcast says Nandan Thakur and colleagues at the University of Waterloo built ORBIT, a 20,000-question dataset of riddles. Each question wraps a short, verifiable answer in four or five clues that narrow the possibilities. The podcast says external search agents re-verified the answers and calls the backwards-built riddles “excellent training data” for search agents. Episode 137 also explores research-agent design, context compaction and memory, and sequential versus parallel search during training.