Comparing NLSQL and ChatGPT-4o Data Analytics Capabilities on US Import-Export Trade Balance Dataset

In this blog post, we will compare the data analytics capabilities of NLSQL and ChatGPT-4o using the same dataset and questions. The dataset used for this comparison is an open dataset from the US Census, containing import-export trade balance transactions worldwide by product groups between 2008-2023.

We began by sending a CSV file containing the full dataset to the latest ChatGPT-4o model. The first question asked was, "What is Canada crude oil import last year?". ChatGPT-4o replied that there was no data for crude oil imports by Canada for the latest year in the provided dataset. However, this was incorrect, as the data was present in the CSV file. We then asked the same question to NLSQL.
NLSQL bot responded, "Thank you for your query. As per the data available, the import of crude oil from Canada to the USA was 92531.63 million USD in the last year." This answer was accurate and could be verified using the CSV file data or generated SQL:

SELECT SUM([trade_balance_usa].[value])
FROM [dbo].[trade_balance_usa]
WHERE year([trade_balance_usa].[year]) = '2023'
AND [trade_balance_usa].[product] = 'Crude oil'
AND [trade_balance_usa].[country] = 'Canada'
AND [trade_balance_usa].[data_type] = 'import'

In this specific test, NLSQL demonstrated its ability to accurately analyze and extract information from the dataset, while ChatGPT-4o failed to provide the correct response. The history of ChatGPT-4o questions can be verified through the provided screenshot. To test NLSQL's capabilities with the same import-export trade balance data source, you can access the open demo using the link below.

NLSQL Open Demo containing import-export trade balance transactions worldwide