🔗 Chaining
By using multiple prompts and at least one ChatPrompt
for orchestration,
you can chain your prompts together. This has many benefits, including:
🗃️ Isolation
By breaking your functionality into separate prompts, its easier to organize your prompts logic into different categories or subtasks, leading to cleaner code that should also be easier to test.
✅ Improved Accuracy
By breaking your functionality into separate prompts, you help the LLM break large problems down into smaller more specific tasks. With the top prompt acting as a router, it can choose to invoke child prompts when needed.
This limits what an single prompt must do, which is important since the more complexity you add to a models prompt, the less accurate it becomes.
💬 🔈 📷 Multi Media
The orchestrator must be a ChatPrompt
to support functions, but child prompts
can be of any type. This is useful because many models today do not support
multiple media types.
You can create a multi-media experience with any model via chaining.