I found the holy grail: Vibe computing + Literate programming = ❤️🔥
Its a bit broken at the moment as I am porting the backend to catch up with o3-mini capability but yeah, it works really well.
observablehq.com/@tomlarkworthy/robocoop
Even better is putting the test suite in the notebook and then you can have it fix the bugs. Which is what I used to write a decompiler for Observable observablehq.com/@tomlarkworthy/observablejs-toolchain, a pretty technical problem that does not permit hallucinations.
📝 Roboco-op: a computational blackboard for efficient human/AI collaboration
Robocoop is a different approach to building an AI coding assistant. In Robocoop, the LLM context is paired with code so it can be hand edited, imported, deleted and forked at any point during development. Its like Chat + Computational Notebook + RAG all wrapped into one. Roboco-op blends http://Observablehq.com|Observablehq.com's reactive notebooks with an open source userspace AI coding assistant. Observable notebooks are a unique coding environment because the code development and runtime state are mixed together. This means Ro
📝 Bidirectional Observable JS Runtime Toolchain
Compilation, source to runtime variable(s) Compilation takes notebook source cells written in
Observable Javascript
and turns them into reactive variables for execution in theObservable Runtime
. A cell is usually compiled to one runtime variable, however, mutable variables are more complicated and are represented as three runtime variables. ObservableHQ does the compilation process as part of the hosted notebook experience but in this notebook we provide a way to do it in userspace. Decompilation, Runt
I am actually against vibe coding coz to me it sounds like a compromise that we are unable to direct AIs properly for coding. So I am much more into test guided AI development, but both massively gain from having a literate programming scratchpad for ideation on technical strategy and goal setting.
From the lab where I’ve been working, here’s an example of a method by which you can steer them
📝 Syntactic and Semantic Control of Large Language Models via...
A wide range of LM applications require generating text that conforms to syntactic or semantic constraints. Imposing such constraints can be naturally framed as probabilistic conditioning, but...