August 2024 Update: Now a solved problem. Use Structured Outputs.

Large language models (LLMs) return unstructured output. When we prompt them they respond with one large string. This is fine for applications such as ChatGPT, but in others where we want the LLM to return structured data such as …

You’re in a new city for a weekend and take a bunch of photographs. When you get home you review your shots and delete some of them. The Keeper Rate is the ratio of shots you keep versus those you throw away. It’s a concept, not a literal metric.

I’ve got a camera on my phone, a …

A collection of film photography shot over April and May. My aim was to experiment with films I’ve never shot before. All pictures were taken an Olympus OM-1 with a 50mm lens.

The appeal of a road to nowhere. The South Island of New Zealand has the Milford Road which brings the traveller to the vista of Mitre Peak and the Milford Sounds. Tasmania has the Gordon River Road terminating at the hydroelectric systems at Lakes Gordon and Pedder. The eighty five kilometre road …

In this blog I share my experience in building a Python REPL augmented with ChatGPT. I explore how the application is built, and speculate on software engineering patterns and paradigms that might emerge in systems built on Large Language Models (LLMs).