The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning | Read Paper on Bytez