The crazy idea I had today was, Oh what fun would it be to write and teach and make videos about whatever I build everyday.
So here we are.
Today, I built an automation for a script that automatically picks up local zoom recording files and uploads them in a shareable way in a notion db.
While doing so, I used quite a few application interfaces.
Windows file system, Notion API, YouTube (Google) API, and the input data coming via zoom saving the recs to a folder.
Amidst all of this, I used openAI in one step to bring all the data together and push it to notion.
Why this decision - because LLMs are perfect for making the non-deterministic → deterministic. It’s the tool when you have a bunch of information but it can’t really talk to each other
Or if it’s just too much work to figure out all the edge cases and if-then-else flows. When writing and maintaining code for a mini task is almost so that it makes the project not worth pursuing → that’s where gpt apis work the best.
I used it to be a link between the knowledge of which tables and databases we use and the zoom file names.
This way, when we add a new client, or when there is any change in where and how we want to manage zoom recordings and how they’re shared - we don’t have to change and rework a bunch of code, we can just simply tell it to the openai system prompt, just as one would to their colleague.
As the cofounder of Nvidia says - one day all code will be natural language, I truly believe him.
Just as easy as it is now to do art, to paint, to make a paper greeting card or to write using a pen - that’s how easy it will be to make digital products that manipulate information.
At least for smaller scale projects, I believe that the natural language code solutions future isn’t that far.
In a way if you see, programming languages have ever been upgrading to be more like natural language. And now we’ve finally taught computers how to better talk to themselves.
This is a new form of leverage we’re unlocking, and I’m excited to see where this goes!
Slowly I’m using such gpt functions in many workflows to make non-deterministic non compatible data talk to each other, reason and output in a deterministic way.
Let’s see which of such solutions impacts things assymetrically.