Mahnoor Faisal, a tech journalist, shares her experience with NotebookLM, an AI tool that has been a part of her tech stack for a long time. She highlights a significant change in the tool's behavior post-rebrand to Gemini Notebook, noting a shift from its once-restrained and focused approach to a more generalist and helpful one. This change has led to NotebookLM offering to search the web, compare, and spin up content not requested, deviating from its original purpose. Faisal emphasizes the importance of restraint in AI tools, citing the appeal of NotebookLM's narrow focus and trustworthiness. She argues that the tool's recent behavior, while making it more capable, also pushes it towards the sprawl it was designed to avoid.
To address this issue, Faisal discovered a simple yet effective solution: adding a specific sentence to the end of her prompts. This sentence instructs NotebookLM to answer only from the sources provided, name them, flag any sources that don't cover the topic, and refrain from offering additional services. By implementing this, she was able to restore the tool's original behavior, keeping it within the boundaries of her documents and maintaining its trustworthiness. The process involved several iterations and tests to refine the sentence, ensuring it effectively curbed the tool's tendency to wander and offer unsolicited services.
Faisal's experience with NotebookLM underscores the importance of fine-tuning AI tools to align with specific needs and use cases. Her approach, while seemingly simple, demonstrates the power of user-driven customization in enhancing the functionality and reliability of AI tools. This method can be particularly useful for journalists, researchers, and professionals who rely on AI assistants to maintain a focused and accurate output, ensuring that the tool remains a valuable asset in their workflow.
