This Is How Reasoning LLMs Really Work

Modern language models don’t just predict the next token anymore — they reason. In this video, we visualize what actually happens inside a reasoning LLMs like OpenAI's o1, o3 etc. Instead of producing an answer immediately, the model explores multiple internal paths, checks its own reasoning, backtracks from dead ends, and only then commits to a final answer. You’ll see how this internal “thought tree” works, why some branches fail, how self-verification improves accuracy, and how techniques like test-time compute and reinforcement learning shape these behaviors. We also discuss why reasonin…