AWS Certified AI Practitioner Practice Exam

What action can help reduce manipulation risks in large language models (LLMs)?

1. A. Create a prompt template that teaches the LLM to detect attack patterns

2. B. Increase the temperature parameter on invocation requests to the LLM

3. C. Avoid using LLMs that are not listed in Amazon SageMaker

4. D. Decrease the number of input tokens on invocations of the LLM

The correct answer is: A. Create a prompt template that teaches the LLM to detect attack patterns

Creating a prompt template that teaches the LLM to detect attack patterns is a proactive measure to improve the model's robustness against manipulation. By designing specific prompts that condition the model to recognize and respond appropriately to potential threats, you are effectively training it, at least in a surface-level sense, to become aware of common attack vectors that might seek to exploit its outputs. This approach enhances the LLM’s ability to discern between benign prompts and those that may lead to malicious outcomes, thus providing a layer of defense against various manipulation strategies that could compromise the integrity of the model's interactions. Such training can foster better responses when facing sophisticated manipulation attempts, aiding in maintaining the reliability and safety of the LLM. In contrast, increasing the temperature parameter can lead to more random and unpredictable outputs, which might make the LLM more susceptible to manipulation, rather than reducing risks. Avoiding LLMs not listed in Amazon SageMaker may not directly address the inherent vulnerabilities and risks associated with the models, as those listed could still have manipulation risks. Lastly, decreasing the number of input tokens might limit the model's context and understanding, which could potentially enhance the risk of misinterpretation or incorrect responses instead of mitigating manipulation.