
Problem Decomposition and LLM Non-Determinism

#165 - Learning to Program in the Era of Generative AI - Leo Porter & Daniel Zingaro
Topic 3 of 10
Published Mar 4, 2024 · Clip 16:19–23:39 (7:20)
The discussion covers problem decomposition as a now-first-class skill in intro programming, since students must break large vague projects into smaller tasks for LLMs. Dan describes his CS-1 projects like analyzing Kaggle datasets, far beyond traditional scope. They then explain why LLMs are non-deterministic, contrasting with the deterministic assumptions of conventional computing, and note this non-determinism is both challenging for lecturers and beneficial since retrying can yield better code. They stress they still teach Python because students must evaluate multiple candidate solutions.