Self-referential prompts are a type of prompting technique that utilizes the model’s own output to further guide its behavior. This can be achieved by feeding the LLM’s generated text back into itself as part of the prompt, or by using the output to create new prompts that are more specific and tailored to the desired outcome.
Guiding the AI’s Approach: Instead of just asking for an output, a meta prompt might guide the AI in how to think about or approach a problem. For instance, asking the AI to generate a response as if it were a specific historical figure, considering how that figure might think or speak.
Reflecting on Outputs: Meta prompting can also involve asking the AI to reflect on or analyze its own outputs. This could include asking the AI why it chose a certain word or concept in its response.
Iterative Improvement: In some cases, meta prompting can be used to iteratively refine the AI’s outputs. After generating an initial response, further prompts can guide the AI to refine, expand, or reconsider aspects of its response.
Chain-of-thought reasoning and Meta Prompting are examples of Self-referential Prompts.
See Also: Chain-of-thought Reasoning, Meta Prompting