Below the Cut Line

At a company where I worked, an impactful prototype developed by a researcher sat on the vine, constantly ending “below the cut line”. Frustrated, he took matters into his own hands and implemented it within the product under a feature branch, even coding in a framework he had little familiarity with, ultimately succeeding at making it work in record time (note: before our era of AI-assisted coding!). Unfortunately that still proved insufficient as the feature never made it into the official product.

A model of reality (or simply knowledge or a theory) within a domain is essentially a “compression” of facts that allows us to calculate in an abstract world, and predict in the real world. A classic example is the ability to explain the motion of a wide range of objects in a wide range of situations with just three Newton’s laws of motion.

Generative AI is the new darling of the AI species, leaving traditional AI, which everyone was enamored of for a decade, a little jealous. Jokes aside, Traditional AI, by which I mean an AI that solves classic tasks such as detection, classification, recognition, segmentation, etc. continues to be important, and still is in relatively early stages of widespread adoption in industry.

We interact with software every day where inputs and outputs are well-defined. And as with gadgets like cars and smartphones, we have come to expect 100% correctness of software. Otherwise you cannot trust that your bank balance is correct, or that the flight reservation you booked last month was actually executed. Correctness is the most basic and minimal expectation we have of software. Correctness being a given, the focus for traditional software development is on other equally important things like time and space-efficiency, reliability, maintainability, scalability, etc.