Code Interpreter might be most useful application of A.I. so far, says Wharton professor

Code Interpreter—a new OpenAI tool for ChatGPT that can run code, work with uploaded files, analyze data, create charts, edit files, and perform math—was released Friday for all subscribers to the $20-per-month ChatGPT Plus service. 

For Ethan Mollick, an early adopter of A.I. and management professor at the Wharton School of the University of Pennsylvania, it might be the most useful, interesting application of the technology so far—and the strongest case yet for a future where artificial intelligence is a valuable companion for sophisticated knowledge work.

“Things that took me weeks to master in my Ph.D. were completed in seconds by the AI, and there were generally fewer errors than I would expect from a human analyst. Human supervision is still vital, but I would not do a data project without Code Interpreter at this point,” Mollick writes in a blog post published Friday. 

With the ability to write code and with access to a large memory—users can upload files up to 100MB—it integrates new capabilities into ChatGPT “in ways that play to the strengths” of large language models, the A.I. technology behind ChatGPT, writes Mollick, who as a researcher was granted early access to Code Interpreter.

“Specifically, it gives the AI a general-purpose toolbox to solve problems,” Mollick writes.

In other words, Code Interpreter elevates the A.I. assistant beyond just generating text responses. It’s valuable to anyone—not just academics and coders— looking to do research. As for the future of work, it’s “a sign of things to come,” Mollick writes.

It also has fewer “hallucinations” because it’s using Python, a versatile programming language used for software building and data analysis, he added. “It closes some of the gaps in language models,” Mollick told Fortune, because the output is not based solely on text. The code is being run through Python, which generates error messages if it’s incorrect. 

Feed it a set of data on superheroes, for example, and Code Interpreter accomplishes the grueling task of cleaning and merging the data “in a quite sophisticated way,” with a “relentless” effort to ensure accuracy (though it always needs to be double-checked, he adds). Users can also go back and forth with Code Interpreter when making visualizations of the data, asking for various changes and improvements

Then, it can perform or recommend a decent analysis, writes Mollick. In this case, it did some predictive modeling, where we can predict what powers a hero might have based on other factors.

“The AI is capable of many other analyses (it is ‘just’ writing Python code, after all) but I was often impressed by its ability to select analytical approaches that made sense,” Mollick writes. 

Code Interpreter’s most striking feature, according to Mollick, is the way it reasons about data “in ways that seem very human.” From the analysis of the superhero data, for instance, it observed that the powers were often visually noticeable because they derived from the comic book medium.

From creating a cute chart to doing regression analysis, Code Interpreter could be the automation people hope for, freeing them from tedious work and giving them more time to focus on “deeper and more satisfying” work. It also democratizes access to complicated data analysis, he told Fortune. 

“Code Interpreter represents the clearest positive vision so far of what AIs can mean for work: disruption, yes, but disruption that leads to better, more meaningful work,” he writes. “I think it is important for all of us to think about how we can take this same approach to other jobs that will be impacted by AI.”