why write code and suffer compilation loop pain when you can prompt an LLM to program LISP for outcomes?

for explainer

if you wanna take the idea for a spin

just load the module you need from outcome

this is super fun

it truely is - now on GitHub too :D

In some way I have done it in my product , and I just soo love this direction. But will we need any software in one year ?

Get the feeling im not smart enough for this shit yet, ill come back in 3 months

You should be able to develop your application within the application itself. Remove this entire notion of GitHub Actions and everything else. Just develop your application.

Well, all that time spent focused on a gfai kernel.. its certainly tempting...

I’m thinking along similar lines but with a smalltalk image with a windowing toolkit

do it!

yessss, when people discover lisp they will go nuts with this. My language for LLMs takes a lot of inspiration from lisps in the fast repl without having to reconstruct state front with some other goodies. It's a superpower

and when the outcome is wrong, how do you even find which part of the lisp did it? you read it yourself or ask the llm again?

i don't know. that's the fun. the more I chat with it, the more it grows from a tiny kernel into anything I want it to be... at a high level: - define properties that must hold true - llm can do anything, including defining new functions - functions and properties are persisted across generations - llm is used less and less as it grows because it's defining deterministic functions.

Right up until the outcome is a stack trace nobody on the team can read 😉

building autocad style REPLs with llms is super fun!