Tencent AI Lab Unveils Persona-Driven Data Synthesis Method with 1 Billion Personas, Achieves Gains on MATH

Jul 1, 2024, 06:23 AM

Researchers at Tencent AI Lab Seattle have introduced a groundbreaking persona-driven data synthesis methodology that leverages a large language model (LLM) to create diverse synthetic data. This method utilizes a vast collection of 1 billion diverse personas curated from web data. The approach has demonstrated significant improvements in performance metrics, notably achieving massive gains on MATH, increasing from 49.6 to 64.9.