I did some more experimentation with open-interpreter today. The first use case I tried was to create, organize and reorganize files. It didn’t generate interesting content, but it was fluent at writing Python code to organize and rename files. When I prompted it to generate a fake dataset, it installed faker and created a CSV with the columns I requested. When I requested it plot those data points, it installed matplotlib and did so without issue.

From there, I moved on to doing some stock analysis. Some coworkers and I have a running joke that the stock price of our company is always lower when the trading window for employees is open. I had open-interpreter do analysis of this, giving it the windows for open trading and prompting it to analysis average and high stock prices both inside and outside the windows. Finally, I had it graph all of these elements together, as it saw fit. I found the results to be impressive, especially given how high level my instructions were.

Open-interpreter writes and runs fluent Python. Do not underestimate how powerful this can be. It gives you what feels like close to an order of magnitude of additional power operating your computer in certain ways, including restructuring the filesystem and analyzing data. I’m eager to try doing more things with it.

Occasionally, it gets stuck and can’t get unstuck, even using the error message output by executing the command. Issues like these are especially annoying if it generates a lot of code, then tries to self-correct and needs to generate it all again.

I wish it were faster. This desire hardly seems fair given that using a language model to write code in this manner feels like superpower, but I found myself itching for it to hurry up. Again, not entirely reasonable, but it’s what I’d want given a “perfect” user experience.

Having not had much opportunity to play with OpenAI’s Code Interpreter/Advanced Data Analysis feature in ChatGPT, I was floored doing the stock analysis. In 5 minutes, the model helped me answer a question I had had but didn’t have the time or motivation to answer myself. The final chart the model constructed with code would have probably taken me a couple of hours of focused work to achieve, and I likely would not have done as good a job.

I’m excited to do more with this tool.