Hi there, @Philip Durbin asked me to share for public, so here we are.
It looks like we're able to "talk" to data directly in natural language and get back verifiable answers from LLMs, with SQL statements. Check this simple dataset:
Screenshot-2023-12-01-at-13.06.32.png
Conversion to SQL table, LLM does recognition of the structure and fill it with data values from the dataset:
Screenshot-2023-12-01-at-13.09.05.png
SQL table is generated out of data, SQL statements provided by LLM to reduce the probability of hallucinations:
Screenshot-2023-12-01-at-13.13.34.png
Last updated: Nov 01 2025 at 14:11 UTC