Stephen Downes

This is a really good article describing a new sort of job role called 'inference engineering'. This is the person who create the layer between the set of weights that constitute the 'model' in an AI system and the application surface where the output is intended to be applied. Here's a really simple of my own design illustrating this: a system, like ChatGPT can autocomplete sentences, but always gets this one wrong: "I'm on my way home from...". That's because it doesn't know where I am. An inference engineer would create the structure that lets ChatGPT figure out my location to fill in the sentence properly. In yesterday's newsletter we saw the (accidental) release of Claude Code's inference layer, including all the tools it can use to interact with the terminal and other systems the user is using. If this interests you at all, definitely read this article.