Meet ChatGPT, the data assistant. — Home

Over a year ago I wrote this blog post showing off the then-amazing analytical capabilities of ChatGPT 3.5-Turbo. To get it working required minimal setup. Setting up python, jupyter, install required packages, etc. And the results were pretty impressive. There were parts where capabilities just weren’t there, so it would instead tell me exactly how to get the data that I asked for. This raises another aspect. So much of it depended on the prompt and the instructions. While writing that blog post I spent some time “prompt tuning” to get the output I wanted.

Almost exactly a year later, OpenAI released its newest multimodal model, GPT-4o. As I was looking at the new capabilities, it occurred to me that this ChatGPT had disrupted itself from a year prior (together with a lot of upstarts in the data analysis business). So, I gave it a try. Below is the entirety of the interaction.

First I provided the dataset

Then after loading, ChatGPT confirmed.

Then here is where the truly mind-blowing things happen. All I provide is the following prompt: Please summarize, analyze and provide relevant charts.

It created relevant charts, summarized and relevant stats from an extremely simple prompt.

From a technological point of view is hard to overstate the incredible pace of development of OpenAI. From a business standpoint if you’re in the AI/ML business is a scary prospect because in a sense: GPT will eat your moat sooner or later. From an end-user perspective GPT-4o is a major knowledge unlock and major force multiplier.

I’m eager to see how this post will compare next year.

Originally published on May 16, 2024.