Adding a badge to your project
title: "DO's and DOnt's with Vibe Coding" date: 2024-01-31
DO's and DOnt's with Vibe Coding
Summary notes of https://www.infoworld.com/article/4029093/9-habits-of-the-highly-ineffective-vibe-coder.html
-
Do's Do experiment with different models. Different LLMs have varying strengths and weaknesses, and you might need to try a few to find the right one for your specific coding problem.
-
Do treat the AI as a search tool. LLMs excel at finding and recalling information from their vast training data, such as obscure documentation or code snippets.
-
Do use the AI for simple, repetitive tasks. The article suggests AI can be helpful for basic programming chores.
-
Don'ts Don't trust the LLM completely. The article warns that LLMs can provide incorrect information and even "hallucinate" non-existent code or URLs.
-
Don't treat the LLM like a "dumpster." Avoid dumping large blocks of code on the AI, as this can be inefficient, costly, and even confuse the model.
-
Don't assume all models are the same. Each LLM has a unique internal structure, parameter count, and training data, which can significantly affect its performance.
-
Don't ignore the costs. AI tools charge by the token, and repetitive requests or large inputs can lead to surprisingly high costs.
-
Don't hand over full control. The article highlights the risk of trusting an LLM with critical tasks, as their inherent randomness can lead to unpredictable and destructive outcomes.
-
Don't expect the AI to "think" like a human. LLMs are clever mimics and excellent at information retrieval, but they are not always good at deep synthesis or providing novel insights.
-
Don't create inconsistent code. The randomness in LLM output can lead to a "patchwork quilt" of different coding styles, making the codebase messy and hard to maintain.