When someone proposes an idea to you, smile and assure them that you understand (yes ...). Express your enthusiasm about it (... yes ...) with an even bigger smile. Then, offer to enhance and extend their great idea (... and ...). You can compare this creativity technique to rubber duck debugging, but always remember that you should not replace a human with an LLM.
The post discusses AI hallucination - when AI generates incorrect information. It explores two main problems: user frustration with incorrect outputs and uncertainty about managing these errors long-term. Using a geodetic network analogy, it explains how AI errors can propagate like measurement errors in surveying, suggesting we need better frameworks for detecting and managing hallucinations.
In this experiment I used AI to automate architecture documentation by testing Aider (an AI coding assistant). After just 5 minutes and 5 prompts, I generated a decent C4 diagram for a Streamlit web application. While not perfect, this experiment shows the promising future of AI-assisted documentation.