A ridiculously stupid software. When it makes mistakes, no matter how you try to correct it, it’s useless. Even when you explicitly point out the issues in the code snippets, it stubbornly insists on messing up perfectly fine code—over and over again! Really don’t know what else to say…
1 Like
I’m getting good results by planning everything with documentation. Otherwise, I feel your pain.
Sometimes, the best strategy is to revert all the changes and re-write the prompt. Be more specific and guide the model more. If a chat starts off badly, it’s sometimes impossible to steer the model in another direction.
As the next person points out, start with a planning session and take a look at the approach presented by the AI. If it’s a good one, make it write a detailed prompt that you can then use with the agent in a new chat to actually make the changes. I’ve found this approach to be much more effective than just trying to one-shot things all the time.
1 Like
Start any conversation by telling it not to do that. Tell it to put in a comment in the code that will prevent it from changing that in the future.
At the smallest disrespect, just go back to the last prompt and change it to guide it where you want, Revert and Continue from that prompt.
Or start with a Thinking model and ask it for a plan at the start of the chat.
Try a different model. I spent 2$ with a MAX model that went off rails and made hundreds of lines of changes, then I restarted the prompt with a single non-premium query and it one-shot the feature without errors in just 0.04$. You are not going to know when you are on the local minima of results/prompt efficiency.
Remember, it’s VIBE coding, catch the waves!
1 Like
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.