IMPROVING MATH SKILLS IN BIG LANGUAGE MODELS

Summary from the TLDR AI newsletter

This research looks into how well LLMs can solve math problems and how different factors affect their abilities. They’ve found that a method called Rejection sampling Fine-Tuning (RFT) helps these models get better at math, especially when we use samples from several models.