Real-World Inquiry Dataset: Bridging the Gap in AI Responses.
During this four-month project, I focused on creating a training dataset that captures the tough questions people often ask AI models but don’t get good answers to. I conducted surveys and interviews to gather insights from real users, identifying over 800 questions that highlight the limitations of AI. Each question was carefully labeled based on its complexity, context, and the type of response users typically expect, covering various areas like healthcare and finance. To ensure quality, I implemented peer reviews and random checks to confirm accuracy and relevance. In the end, I compiled around 1,200 unique questions, which will help train AI systems to better understand and respond to the real challenges users face when seeking information.