Why Superintelligence Might Still Need Humans
A comic strip exploring why even a superintelligent AI might still need humans around.
ReadA comic strip exploring why even a superintelligent AI might still need humans around.
ReadContent provenance is the attempt to put the signature back — to answer the question the internet made us stop asking: who actually made this?
ReadI asked Claude Code to add markdown files and push. It ran 30+ commands and diagnosed a broken CI pipeline I didn't know about.
ReadAI collapsed the cost of iteration. The only remaining explanation for a mediocre product is a mediocre set of decisions.
ReadAn interactive AI resume that lets people ask questions about my work instead of reading a static PDF
ReadI built a command-line coding agent in very few lines of Python to understand what these systems actually are
ReadI built a a portable work state CLI tool to easily switch between LLMs
ReadLet's take a local manifold walk to understand why small models are sometimes better, more resource responsible and won't burn a hole in your pocket.
ReadI'm running into rate limits every hour or so into using claude code. I analyzed one session: < 3% of my AI bill went to intelligence. The rest? Regurgitating stale and bloated context
ReadI was fixing a bug for my newborn's app and ChatGPT had a complete brain-body disconnect: It kept reasoning 'don't do the thing' → did it → 'oops I did this' → redo → 'oops I did it again'
ReadDemos are easy. Production is hard. If you want internal knowledge search that actually works, you're not 'adding RAG'—you're building a pipeline where the LLM is the easiest part.
ReadFrom Evals 101 to Production Grade systems.
ReadProduct lessons: atomic desires, two way doors, and future complexity backseat
Read