What to Know About ChatGPT's New Code Interpreter Feature – The New York Times

Supported by
Graphs, maps and data analyses? Now ChatGPT can do even more.

Reporting from San Francisco
ChatGPT, the artificial-intelligence-powered chatbot made by OpenAI, has wowed the world in recent months with the text it can generate. Now the chatbot is delighting users anew by creating charts and maps and turning images into videos.
That’s because of a new feature that OpenAI rolled out to ChatGPT Plus subscribers last week, called code interpreter. Here’s what to know about it.
Code interpreter allows ChatGPT to analyze data, create charts, solve math problems and edit files, among other uses. It also supports uploading and downloading files, which was not possible in ChatGPT before.
Code interpreter became available Thursday to subscribers of ChatGPT Plus, a service that costs $20 a month. Similar add-on features, which give ChatGPT users access to third-party services such as Expedia and OpenTable, are available to subscribers only.
When people ask ChatGPT a question, the chatbot guesses an answer based on a technology called a large language model, or L.L.M., which predicts the next word in a sequence.
But when code interpreter is enabled, ChatGPT writes and runs a piece of computer code to find the answer, OpenAI said. That lets the chatbot accomplish new tasks that it didn’t do before, such as performing complicated calculations and generating charts based on data that a user uploads, which are all completed by code.
Some argue code interpreter reduces the chances of inaccuracies, a common complaint about L.L.M.s.
“The code objectively does something right,” said Ethan Mollick, an associate professor at the University of Pennsylvania, who tested code interpreter for two months before it was released.
The most common use of code interpreter is data analysis.
With a prompt like “tell me what is interesting about the data,” ChatGPT can look through a user’s data, such as financial, health or location information, and produce insights about them. Financial analysts have found ways to use the tool in their work, such as analyzing stock prices and preparing a budget.
Researchers have also started using the tool for creative data visualizations. Dr. Mollick said he had recently used code interpreter to create a three-dimensional chart of the Billboard Hot 100 list and make an animated map of every lighthouse in the United States. Some people have also used code interpreter to convert the formats of files, such as turning images into videos or PDF documents into pictures.
Arvind Narayanan, a professor of computer science at Princeton University, cautioned that people should not become overly reliant on code interpreter for data analysis as A.I. still produces inaccurate results and misinformation.
“Appropriate data analysis requires just a lot of critical thinking about the data,” he said.
Yiwen Lu is a technology reporting fellow based in San Francisco. More about Yiwen Lu