Hey narush, I'm the author of PandasAI. The way PandasAI works is slightly different. It's not the LLM that edits data, but a python script. This is in my opinion the best approach, as it gives a predictable result.
There are still some potential issues, including hallucinations, that we are taking care of, but all in all we hope that this could represent an advancement in the way we interact with data.
Hey SilverBirch, I'm the author of PandasAI. In terms of the naming, that's a very good point, and I must confess I didn't even consider that when I published the repo. It just made sense to me, since the goal was to add AI on the top of pandas, to call it "PandasAI" (pandas + AI). For it was more of a tribute to the library, but I totally get your point, and I thank you for pointing it out.
I'm trying to reach out the pandas founder, and I'll definitely change the name of the repo if he prefers to.
As of the quality of the output, we are constantly working on improving the quality of it. It provides good enough results in most of the cases, but it still suffers from some hallucinations, although we have already fixed many either programmatically or with prompt engineering.
As of the dataframe, we don't pass the whole df, but only the head and 5 columns with shuffled data (also redacting the sensitive ones).
Any questions, feel free to ask :)