🧠 LLMs can be trained to improve their math skills, particularly multiplication, through an iterative system message optimization process. 🧮
🚀 First, a dataset of multiplication problems is generated, and a baseline accuracy is established with a chosen LLM.
📝 Next, a larger LLM is used to generate a system message that provides a step-by-step approach to solving the problems.
📈 This system message is then refined iteratively based on the smaller LLM’s performance, with the larger LLM analyzing results and suggesting improvements. This process, combined with experimentation of LLMs and detailed feedback, can significantly enhance the mathematical capabilities of LLMs.