I have complex AI instructions (~3000+ words) covering workflows, examples, rules, and formatting requirements. The model seems to get confused. It is not following the formats and rules provided at certain cases in its output.
What’s the best practice?
- Keep everything in one large instruction file.
- Break into main instructions that covers the workflow and keep other instructions in different files
Which approach gives better model performance and consistency? Any recommended instruction length limits?
Using GPT-4o via API.