---
title: "Producing Charts with AI"
description: "Discover how AI code interpreters are revolutionizing data visualization, reducing chart creation time from 20 to 5 minutes while simplifying complex statistical analysis."
categories: ["AI","data analysis"]
keywords: ["AI data visualization","ChatGPT Code Interpreter","statistical analysis tools","data analysis software","R programming","SaaS analytics","machine learning for charts","Tomasz Tunguz","AI in venture capital","data science automation"]
ai_summary: "Explore how AI code interpreters like ChatGPT are transforming data visualization and speeding up chart creation."
date: 2023-07-17
lastmod: 2026-07-31
canonical_url: https://www.tomtunguz.com/data-analysis-gpt/
author: "Tomasz Tunguz"
---

Last Monday, I published a chart that looked like this. It shows the standard deviation in Series A round sizes over time. I needed about 20 minutes using [a language called R](https://tomtunguz.com/how-to-make-pretty-charts/) to create it. 


![image](https://res.cloudinary.com/dzawgnnlr/image/upload/aun7pnnnnlnagfba8zss.png)

Then I tried to replicate it using [ChatGPT3's Code Interpreter feature](https://chat.openai.com/share/2117466e-8d83-411a-98d3-5556d8992e4d) & I spent less than 5 minutes duplicating it, most of the time waiting for the computer (see video at the end of the post).

![image](https://res.cloudinary.com/dzawgnnlr/image/upload/tdr8smjqbai2pofr6dyb.png)

There's a lot to like about using Code Interpreter.

First, I can describe my ideal chart in English. When ChatGPT finds formatting foibles in the data, it fixes them. My columns contained extra spaces.

Second, the computer determines which libraries to use & installs them, so I stop worrying about my environment. 

Last, I can perform statistical analysis like testing for significance, Anova, & correlation just by asking. The robot will explain the test, the rationale for the test, & conditions for when the test works. I can forget about syntax arcana of a particular [student T-test library](https://en.wikipedia.org/wiki/Student%27s_t-test).

For me, this is the power of LLMs. Analysts operate at a higher plane of abstraction & compress the amount of time to achieve an output.

It's not perfect. I spend most of the time waiting for the computer to return after a query. It's unclear how to replicate or share the analysis if I needed to because the session doesn't store the data file & it's not linked to Github. 

But, these features will surely be implemented in the future by SaaS startups.

Here's the video of my session with ChatGPT.
{{< youtube 9G3hUzlJG-U >}}