A Trip Down the GPT-4 Code Interpreter
In the last post, Swimming in the Blue Ocean, I mentioned how I used the GPT-4 code interpreter to reconstruct the value curves found in the book ("The Blue Ocean Strategy"). This post is a behind the scene look at how I generated these, as well as some thoughts and first impressions.
We'll aim to recreate the strategy canvas comparing Southwest Airlines to the airline industry and car travel. You can then use the same techniques to create a value curve for your own product and industry.
The final output looked something like this:
This post is a 6-8 minute read and for Paid Subscribers only.
If you are interested in reading the full post, please upgrade now:
As a brief refresher, the GPT-4 Code Interpreter is a new addition to Chat-GPT (also only for plus subscribers). This can, in effect, run, execute, and even debug! Python code in a sandboxed environment and give you back the results. The first time I used this, it blew my mind. I just gave it a prompt, and a few seconds later, I had an image with near-perfect results. Not only did it take a CSV and generate the code I wanted, but it also spotted an error in the code (one of the columns had extra whitespace), debugged and fixed the error by adding code to trim the whitespace, and correctly generated the output without any intervention from me. That's neat.
I've been using Chat-GPT to generate code here and there before, but this was the next level. Before, it was like having an intern; now, it's like the intern got a promotion and is more like a junior employee. I've only used the code interpreter to generate charts and perform data processing tasks. But there are loads more applications (image processing, code editing, etc.), and it's likely only going to get better.
It all runs in a sandboxed environment (that basically means some kind of isolated virtual instance), and it seems to have a (rather short) time limit before the instance is recycled. So you can't use this as your own private server, but as a personal assistant, the possibilities are endless! There's probably an API as well, but I haven't messed around with it yet.
The time limit was a bit annoying if I'm being honest, but I'm assuming this will get better. It's obviously understandable to not want to have all these instances lying around. Currently, it takes longer to restore the state after a time-out than to run the prompt from scratch, so keep that in mind.
Here's a step-by-step guide on how I used it to generate these charts. This is valid as of early August 2023. The steps might change if you try it later on.
I haven't done a “tutorial” per se in a while, so this should be fun!
Here we go …
Keep reading with a 7-day free trial