Imagine needing realistic but fake data—for testing software, training AI, or protecting sensitive information. Creating convincing synthetic tabular data, the kind found in spreadsheets and databases, has always been a challenge. Traditional methods often struggle to capture the intricate relationships between different data points. But what if we could leverage the power of large language models (LLMs), the brains behind AI chatbots, to generate this data? That's the idea behind AIGT, a new technique that uses AI prompts to create incredibly lifelike synthetic tables. AIGT works by feeding metadata—information about the table like column names and descriptions—to an LLM as a prompt. This gives the LLM context, allowing it to generate data that accurately reflects the real-world relationships within the table. But what about massive datasets that exceed an LLM’s capacity? AIGT cleverly partitions large tables, generating synthetic data piece by piece before seamlessly stitching it all back together. Tested on a wide range of public and private datasets, including real-world financial data from Alipay, AIGT consistently outperforms other leading synthetic data generation methods. This breakthrough opens exciting doors for improved data privacy, more robust machine learning models, and better software testing. While AIGT's processing speed can be improved, this research showcases the potential of LLMs to tackle complex data challenges in a clever and effective way.