Matt Welsh, CEO and co-founder of Fixie.ai, founded by Google and Apple natives, believes the traditional idea of ​​”writing a program” has disappeared. Most software will write artificial intelligence systems.

Where a “simple” program is required, it will all be AI-generated, not handwritten, and at best, humans will be assigned a controlling role, Welsh says: “In the future, students will not need to learn mundane skills like adding nodes to a binary tree or learning C++ code. This type of education will be as outdated as teaching engineering students to work with a logarithmic ruler.”.

With a few keystrokes, future engineers will launch a model instance with four quintillion parameters, where the entire volume of human information is already encoded and ready to perform any task set before the machine. A big part of the smart job of getting a machine to do what it’s supposed to do will be to find the right examples, the right data to train on, and the right ways to evaluate the learning process.

Strong enough AI models that can generalize through multi-step training will only need a few good examples of the task. Large, man-made datasets will no longer be needed in most cases.