Artisanal Computing: Handcrafted Ruby Execution for the Modern Age
summary
A human-powered VM runs MRBC bytecode inside a Google Sheet by translating 119 mrb_opcodes down to a five-instruction human-friendly ISA, exporting JSON to a spreadsheet renderer, and driving execution with crowdsourced workers on Amazon Mechanical Turk; a headless LibreOffice process acts as a virtual HPU for end-to-end tests. Training and latency blow up costs (tasks cap at $10 so programs must be segmented), MTurk enforces budgets causing a halting problem workaround, and LLMs are used iteratively to learn dense MRuby/Prism internals and land two PRs to Prism.
AI-written summary · about
chapters (11)
- 0:00Handcrafted Ruby execution
- 1:48Spreadsheets vs traditional VMs
- 3:16Human-powered virtual machine
- 6:50Instruction set breakdown
- 9:16Recursion in spreadsheet VM
- 13:08Testing human-executed VM
- 15:36Cloud human processors
- 19:55Mechanical Turk limitations
- 21:39Contributing to Rails
- 23:24AI scaffolding Ruby internals
- 25:10Porting M Ruby to Prism
