Bidirectional search helps language models solve hard reasoning problems

Original: How do we get LLMs to solve hard reasoning problems that the base LLM can barely solve?

Source: x.com ↗