Code interpreter is a game changer


In the near future, we might see an influx of excited discussions regarding utilizing code interpreter for various tasks. I'm not going to do that in this post, although I do provide an illustration of how code intepreter can be used to solve a complex numeric problem. My focus instead is why code interpreter signifies a pivotal shift, one that has been taking place quietly for some time but is now more clearly visible.

Why is code interpreter a game changer?

I have suggested for some time now that LLMs can be seen as intelligent API gateways and code interpreter, I think, illustrates this particularly vividly. This has profound implications for the way in which, for example, analysis is performed. Why would anyone now invest in analysis services when a suitably set up LLM (i.e., ChatGPT and code intepreter) can deliver similar, if not superior, results at no cost? Yes, we still need to address scale and deployment questions, but analysts, consultants, auditors, accountants, data scientist, and their clients should take note. The reason being, they can now access sophisticated analyses on just about anything within the ChatGPT environment, free of any additional charges, except for the ChatGPT subscription itself.

What is code interpreter?

Code interpreter is perhaps best described as a personal data analyst. It can read uploaded files, execute code, generate diagrams, statistical analysis, and much more. I expect it will take the community some time to fully chart its potential.

To activate code interpreter:

In ChatGPT Plus, on bottom left click on your name > Settings > Beta features > turn on Code Interpreter.

Using code interpreter to solve a complex numerical problem

In the screen shot below you can see how I use code intepreter to import 533,860 rows of OHLC tick price data from the NYSE TAQ. Code intepreter handles this data volume without issue. I ask it to provide descriptive statistics for this dataset, plus an augmented Dickey-Fuller test statistic and p-value, replicating some of the initial data analysis I perform in my earlier post Using a GPT to predict volatility. Due to the size of the file, I instruct code interpreter to use the first 5000 rows only for the analysis.

Using code interpreter to solve a complex numerical problem

Here are the results:

Using code interpreter to solve a complex numerical problem

You can compare code interpreter's results to the results I get in my earlier post Using a GPT to predict volatility.

Source code

Source code as always can be found on my GitHub.


openai.com, ChatGPT plugins. ChatGPT plugins.

© 2023 johncollins