Bounty: Diverse hard tasks for LLM agents — LessWrong