Listen to the podcast:
The Power of
Trustworthy Content: Enhancing Generative AI Applications
Disclaimer: This podcast is automatically generated using AI
Generative AI has rapidly gained traction in spaces like customer support and personal productivity. So far, adoption in education remains limited. Key barriers include concerns about responsible use, low perceived value, and lack of trust in AI-generated responses.
In this blog post, we’ll explore the promise of using trustworthy content in improving Gen AI outcomes, and why it’s essential for learner-facing applications in education.
What do we mean with Trustworthy Content?
Trustworthy content is accurate, expert-validated and relevant to its intended purpose. It originates from credible sources, reflects the latest knowledge in the field, is unbiased and comprehensive, and, in education, aligns with curricula or course objectives.
When generative AI applications are trained or fine-tuned using trustworthy content, they produce responses that are accurate, contextual, and aligned with users’ needs.
Why Does Trustworthy Content Improve Gen AI Outcomes?
Generative AI systems thrive on the quality of their input data. Large Language Models from OpenAI, Anthropic and others, while powerful, can sometimes generate inaccurate or nonsensical outputs. Trustworthy content elevates performance in three ways, namely by improving accuracy, boosting user trust and fostering critical thinking skills.
Improving Accuracy: Models grounded in validated information significantly outperform the out-of-the-box ones. For example, Wizenoze conducted research using OpenAI’s SimpleQA dataset and showed that responses grounded in validated content achieved an impressive 91% accuracy, compared to 30-40% without content. These findings hold for cases when a clear fact base is available to support LLM output. LLM providers like Perplexity have looked into the direction of Retrieval Augmented Generation (RAG) to improve accuracy by extending the LLM with search capabilities. This approach with unvalidated search results, yields ~77% accuracy – an improvement, but probably not ‘good enough’ for Education.
The text continues below the image.
Boosting User Trust: In a qualitative user research Wizenoze conducted, users indicated they are more likely to trust and rely on AI systems when clear references & citations are added to the responses. User feedback underscores this: “Wizenoze’s tutor answered questions thoroughly, I like that it included the sources,” said one educator during a field experiment. Another educator mentioned that “the links & resources were like a cherry on the cake.”
Fostering Critical Thinking Skills: Without proper grounding, Gen AI applications risk perpetuating misinformation, particularly in a sensitive area like education. Grounding applications in trustworthy content not only reduces the risks of inaccurate outputs but also fosters critical thinking by encouraging learners to evaluate the reliability of sources.
When is Trustworthy Content Particularly Important in Education?
In education, the stakes are especially high. Few people would argue that 40%-accurate Generative AI applications are good enough.
The SimpleQA benchmark research shows that for factual question – answering type interactions, adding validated content gives a big accuracy improvement. Based on these findings, we foresee adding trustworthy content to have most impact in K-12 Social Sciences courses like History and Sociology. Especially in cases when the AI application is learner-facing (such as personalized tutors), there is a big need for accurate, relevant, and well-contextualized information. If students rely on these tools for clarity on complex topics, incorrect information can lead to confusion and misunderstandings.
The text continues below the image.
Conclusion: The Future of Gen AI is Grounded in Trust
Trustworthy content isn’t just a technical requirement—it’s a moral imperative for developers and organizations building Generative AI applications. In high-stakes domains like education, grounding AI in validated, high-quality data ensures that these tools empower users while avoiding the pitfalls of misinformation.
As we continue to push the boundaries of Generative AI, let’s prioritize accuracy and reliability. Big LLM providers will pursue higher accuracy with new flagship models, but one could argue that doesn’t fire educational companies from their duty to make meaningful improvements. Because in the world of education, 40% accuracy simply isn’t good enough.
