The current generation of Large Language Models have some impressive capabilities. But they don’t seem like good tools for data wrangling:
A natural language/chat interface is a rather ambiguous, clumsy and slow way to describe how you want data transformed.
They generally can’t handle thousands (let alone millions) of rows of data.
They might hallucinate a solution.
You can never be quite sure what they have done.
But we have found that Large Language Models can be useful for finding the right transform. They have clearly crawled all of our documentation and the forum. So they can give pretty impressive answers to questions like ‘What Easy Data Transform transform should I use to calculate the running total of a column of numbers?’:
By the way, I use since a while AI when I need in filter or replace transformation REGEX criteria, I have no real knowledge on REGEX, but AI gives me feasible expressions based on the description I did in my words.
Good remark Mr Andy. That is why I prefer to use (in a first phase) EDT in order to generate clean and well structured data then, in a second phase, AI prompt for python + streamlit.
