New series testing if/how large language models convert questions about a dataset into code that runs on-the-fly to analyze it, all web and client-side
Large language models (LLMs) have proven exceedingly powerful for tens of applications, some not even expected upon design, and researchers (and plain users) continuously unlock new capabilities -and limitations, of course. But there’s one specific pressing question about LLMs that I’ve had ever since GPT-3 came out for programmatic use around 2 years ago:
Can LLMs help with data analysis in ways that make users focus on the science or engineering problem, abstracting from any specific maths, algorithms or programming skills required?
Or, poised in a different, more personal and straightforward way:
Can I ask an LLM questions about a dataset with my own words and have it interpret these questions with the maths or scripting required to answer them?
Short answer: Yes, but…
Long answer: read on and stay tuned to my upcoming articles, where I will unveil quite a lot of power, and some limitations, of LLMs for data analysis. Faithful to my style, with code and web apps for you to try online.
This is something I’ve wanted to do for a long time since I started working with GPT-3 programmatically around a year before the…