I tend to think of AI coding more like a tool and an advanced search engine. It closes the loop of searching for info, adjusting it to my needs, and writing the code. The trick (as this article states) is that you need to give clear instructions, and have the knowledge about what a good solution might look like to get anywhere near a good result. Also clearing or starting new prompts for smaller problems in the code might be helpful because the bigger the history gets, the more the model might be confused about what you want it to solve.
>On Slop and Accumulation
From personal experience I've learned to prompt to add just basic features so its minimalistic and serves the job, even vague prompts like these one liners work "Make sure we are only adding necessary files, You'll get more points for lesser files/lines of code and your goal is to maximize the points you get"
2 comments