Failure patterns, evaluation practice, and diagnostic method from engagements on LLM, RAG, NLP, and agent systems already running in production. Interviews, podcasts, and articles published elsewhere are collected here too.
TypeSafe’s Jev returns an answer with a confidence score. On one task in an independent study it was at least 0.9 confident on 78% of items and wrong on 62% of them. Here are the numbers, including the ones in TypeSafe’s own cookbook.
Read more →
Microsoft Word keeps almost none of a document’s formatting next to the words, so an agent editing your document can’t see the styling it is breaking. Here is the routine I built to catch it, and how to install it.
Read more →
Coverage of the “Teaching with AI: Embrace, Adapt or Resist?” panel at USF Tampa, where I argued that AI should not quietly absorb the work junior employees learn from.
Read at St Pete Catalyst ↗
The AI models I use every day started writing in a way I could barely read. The pattern has a name: asyndetic parataxis. Here is how to identify it, and the prompt that stops it.
Read more →
A live interview on FOX Tampa Bay Live. As students head back to school, I break down what AI should and should not be used for in school work.
Read at FOX 13 Tampa Bay ↗The first notes are being written. In the meantime, tell us what you're working on and we'll give you a senior technical read on the behavior you're seeing.