Hi everyone,
First-time poster here, and I’m just starting to explore the deeper capabilities of prompt engineering in Google AI Studio. I’ve hit a wall with an issue that feels like a classic rookie mistake, and I’m hoping this amazing community can offer some insight.
My Goal:
I’m trying to create a sophisticated, multi-part persona prompt. The idea is to have a base persona (Part A) and then dynamically “merge” it with a specific instruction set for task logic (Part B).
For Part B, I’ve developed a simple system I’m calling FELLAS (short for Few-shot Example Logic & Language Alignment System). It’s essentially a block of examples that guides the model on how to format its response.
The Problem:
Both parts work perfectly on their own. However, the moment I combine (or merge) the FELLAS instruction block with the base persona, the model enters what I can only describe as an “unlimited shake” loop. It starts generating nonsensical, repetitive text based on the last example in my prompt, and it never stops.
Here’s a simplified version of my prompt structure:
### PART A: Base Persona ###
You are a helpful assistant. You are cheerful and concise.
### PART B: FELLAS Instructions ###
Here is how you format your answers.
Example 1:
User asks for a capital city.
You answer: "The capital is [City]."
Example 2:
User asks for a fun fact.
You answer: "A fun fact is [Fact]."
Now, answer the user's next question based on the persona and the examples.