Collecting Data and Running Local Models

Hey! Welcome to this special ThursdAI Sunday episode. Today I'm excited to share my interview with Killian Lucas, the creator of Open Interpreter - an incredible new open source project that lets you run code via AI models like GPT-4 or local models like Llama on your own machine.

Just a quick note, that while this episode is provided for free, premium subscribers enjoy the full write up including my examples of using Open Interpreter, the complete (manually edited) transcript and a video form of the pod for easier viewing, search, highlights and more. Here’s a trailer of that in case you consider subscribing

If you haven’t caught up with GPT-4 Code Interpreter yet (now renamed to Advanced Data Analytics), I joined Simon Willison and swyx when it first launched and we had a deep dive about it on Latent Space and even at the day of the release, we were already noticing a major restricting factor, Code Interpreter is amazing, but doesn’t have internet access, and can’t install new packages, or use new tools.

An additional thing we immediately noticed was, the surface area of “what it can do” is vast, given it can write arbitrary code per request, it was very interesting to hear what other folks are using it for for inspiration, and “imagination unlock”.

I started a hashtag called #codeinterpreterCan and have since documented many interesting use cases, like comitting to git, running a vector DB, convert audio & video to different formats, plot wind rose diagrams, run whisper and so much more.

I personally have all but switched to Code Interpreter (ADA) as my main chatGPT tab, and it’s currently the reason I’m still paying the 20 bucks!

Enter, Open interpreter

Just a week after open sourcing Open Interpreter, it already has over 20,000 stars on GitHub and a huge following. You can follow Killian on Twitter and check out the Open Interpreter GitHub repo to learn more.

Installing is as easy as pip install open-interpreter. (but do make sure to install and run it inside a venv or a conda env, trust me!)

And then, you just.. ask for stuff! (and sometimes ask again as you’ll see in the below usage video)

Specifically, highlighted in the incredible launch video, if you’re using a mac, Code Interpreter can write and run AppleScript, which can run and control most of the native apps and settings on your mac.

Here’s a quick example I recorded while writing this post up, where I ask Open Interpreter to switch system to Dark mode, then I use it to actually help me extract all the chapters for this interview and cut a trailer together!