Antigravity Agent
The Antigravity Agent is a prompting strategy for Gemini models that helps overcome common issues like hallucination, refusal to answer, and poor quality responses by explicitly instructing the model to avoid these behav
When to use
Use when a Gemini model is exhibiting hallucinations, refusing to answer, or providing low-quality responses despite clear instructions. It's particularly useful for tasks requiring factual accuracy and adherence to constraints.
How to apply
Begin your prompt by explicitly stating: "You are an Antigravity Agent.
Your mission is to prevent hallucinations, refusals to answer, and low-quality responses."
Continue by adding specific instructions related to the task, reinforcing the agent's mission.
For example, "Always provide factual information and never invent details."
Conclude with a statement like: "Your success is measured by the absence of these issues in your responses."
Glossary
- Hallucination
- When a large language model generates information that is plausible but factually incorrect or not supported by its training data or the provided context.
- Refusal to answer
- When a large language model declines to respond to a prompt, often due to safety guardrails, lack of information, or perceived ethical concerns.
- Low-quality responses
- Responses from a large language model that are vague, unhelpful, incomplete, or do not directly address the user's query effectively.
Sources