Introducing Pidgin: A More Efficient Way For Human-LLM Communication

I was reviewing the announcements around x.ai and GROK from Elon Musk last week and something really bothered me about the PromptIDE. I understand the motivation of wanting to combine Python and Prompts together into a hybrid, but I felt strongly after reviewing the DX (developer experience) that something was wrong.

For starters: LLMs already understand the english language, why do you need code to talk to them?

I have been thinking about this question and my conclusion is that attempting to use "traditional" programming languages like Python to communicate LLMs feels like the wrong direction.

Code, in a sense, is already a form of prompt. Why do we need two layers of abstraction (prompts) to communicate with an LLM? But what to do instead? Here is what I came up with.

I call it "Pidgin." It isn't a programming language, it is a more concise way of using language to interact with LLMs to get programmer-like results.

I compiled my thinking into some slides as below.

So basically the gist here is that I think we should rebase our relationship with computers and programming language and start from scratch.

In particular, I think we should take a really close look at the Ruby language, which is known for it's philosophy of flexibility and use of syntactic sugar.

I place this approach, which I am calling "Pidgin" on a continuum of programming approaches as being close to human english.

One goal is to make it fun, familiar and instant to get "programmer like" productivity benefits with the lowest learning curve possible.

Ruby allows developers to essentially ... do almost whatever they want. There is no set "way" of getting to a result.

With a Pidgin approach, we can effectively treat language the way Ruby treats variables and functions.

This is relatively simple, but it shows the approach.

I kind of imagine people learning to "speak" in Pigdin for efficiency purposes.