Christian Gudrian<p>Recently I talked to a close friend of mine, who began <a href="https://social.tchncs.de/tags/VibeCoding" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>VibeCoding</span></a> not long ago. He isn't that much into programming and just wants his ideas to become reality. The tasks he wanted to implement were quite simple, so he was very happy with the quick progress he experienced with having an <a href="https://social.tchncs.de/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>LLM</span></a> generate the code for him.</p><p>I can relate to that to a certain extent. Yes, for simple, tutorial level tasks LLMs generate code, that works. Most of the time. As soon as things get a little more complicated or leave the beaten path, they drop out and you enter the stage of <a href="https://social.tchncs.de/tags/DoomPrompting" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DoomPrompting</span></a>, where you constantly confront the LLM with its faulty output, whereupon it apologises, promises to correct itself, and yields a new solution – which turns out to be broken in a different way.</p><p><a href="https://social.tchncs.de/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> <a href="https://social.tchncs.de/tags/GenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>GenAI</span></a></p><p>1/2</p>