The Good: GPT Builder is a great idea overall and will only get better. The Bad: GPT builder is the worst coworker; It will constantly try to erase instructions and write nominal weak instructions i nits place even when specifically told not to change the instructions or directed to a quoted sesction to change. after days of a variety of isntructions and "Create" work experiments, it really is miserable at contextual goals of the builder. Especially if you want to embed any code or blocks of code to include with output. For example if you wanted to include python snippets etc as a template for output, it will constantly just delete code of any sort. I understand the knowledge attachments but in testing it definitely reads better and operates better from the instructions. The second issue is that the preview output and the test page output will reliably do what they do but NOT the same, which is super frustrating and burning ChatGPT 4 uses up as well with many errors and errored streams STILL COUNT against use(thats not great from this side of things). If you use the interpreter or dalle3 option in the builder you run into instances where Dalle3 starts to generate with zero verbage to indicate image generation and the interpreter version will cause constant errors in the output stream to the point of nearing unusable.