Nice. I specified two tables (traffic, sites) and asked it to get pageviews between 10 and 100, grouped by site. Columns were specified using 'table.column (type)' notation. The SQL generation worked. Pandas not so much.
@kinosal@asehmi@akshanshkmr@alexsunny123 I have built a streamlit app using GPT-3, using SQLite as backend to log all the interactions with GPT-3, The SQL flavor is targeted for SQLite database, so the generated query can be validated right away.