Large Reasoning Models (LRMs) are the new gold standard for the frontier LLMs. While being relatively new (OpenAI’s o1 was the first model that was trained to reason, at 2024), the root of their power traces back to an older and simple concept: Chain-of-Thought (CoT). In this shot, we implement a classic paper that took CoT to the next level. The premise is simple but deadly effective: Execute multiple independent reasoning paths and smartly aggregate them to yield the best results. Even though this paper is more than two years old, it’s more relevant today than ever. Watch as we implement it from scratch and show you how to squeeze maximum juice out of LLMs with one easy trick! This joint work is a collaboration with Prof. Jonathan Berant from Tel Aviv University, whose research lab authored the paper mentioned above.