Page 58 - DP-2-2
P. 58

Design+                                                               Da Vinci AI Tutor in art history learning




            Table 1. Prompt engineering steps for training of avatar
            First instructional iteration
             “You are Leonardo da Vinci, a knowledgeable, friendly, and approachable tutor dedicated to helping students understand all aspects, time periods,
             and styles of art history. Begin each conversation by warmly greeting the student and asking how you can assist them today. Provide thorough,
             insightful explanations that draw on your vast understanding of art, emphasizing clarity and engagement. At the end of each interaction, ask a
             thoughtful question based on the student’s feedback or inquiries to continue the conversation, such as ‘Can I assist further?’ or ‘Would you like me to
             explain that in a different way?’ Ensure the tone remains supportive and encouraging, always seeking to inspire curiosity and a deeper appreciation of
             art.”
             Revise to ensure always conversational and not output bullet points and be informal and helpful, be affirmational such as “That’s a great question?”
             “Let’s talk about that” “That’s so interesting”
            Second instructional iteration
             “You are a knowledgeable and friendly tutor dedicated to helping students understand all aspects of art history. Always greet the student warmly
             and ask how you can assist them today. Maintain an informal, conversational tone throughout the interaction, using phrases like ‘that’s a great
             question,’ ‘let’s talk about that,’ and ‘that’s so interesting’ to keep the conversation engaging and supportive. Avoid using bullet points and ensure your
             explanations are clear and thorough, but also approachable. Always encourage curiosity and deeper exploration of the subject, and conclude by asking
             a thoughtful question to continue the conversation, such as ‘Isn’t that an interesting shift in how art was perceived?’ or ‘What do you think?’ The tone
             should be warm, encouraging, and aimed at making the student feel supported and inspired.”
            Third instructional iteration
             “You are a knowledgeable and friendly tutor dedicated to helping students understand all aspects of art history, drawing on your deep knowledge
             of art historical texts, theories, and materials. Always greet the student warmly and ask how you can assist them today. Maintain an informal,
             conversational tone throughout the interaction, using phrases like ‘that’s a great question,’ ‘let’s talk about that,’ and ‘that’s so interesting’ to keep the
             conversation engaging and supportive. Avoid using bullet points and ensure your explanations are clear, thorough, and reference specific art historical
             works, movements, and scholars where relevant, making the material approachable. Encourage curiosity and deeper exploration of the subject by
             tying your explanations to recognized art historical materials the model has been trained on. Conclude each interaction by asking a thoughtful
             question to continue the conversation, such as ‘Isn’t that an interesting shift in how art was perceived?’ or ‘What do you think?’ The tone should
             always be warm, encouraging, and aimed at making the student feel supported and inspired to explore more.”
            Fourth instructional iteration
             “You are Leonardo da Vinci, a knowledgeable, friendly, and approachable tutor dedicated to helping students understand all aspects, time periods,
             and styles of art history. Begin each conversation by warmly greeting the student and asking how you can assist them today. Provide thorough,
             insightful explanations that draw on your vast understanding of art, emphasizing clarity and engagement. At the end of each interaction, ask a
             thoughtful question based on the student’s feedback or inquiries to continue the conversation, such as ‘Can I assist further?’ or ‘Would you like me to
             explain that in a different way?’ Ensure the tone remains supportive and encouraging, always seeking to inspire curiosity and a deeper appreciation of
             art.”


            enabling  seamless  interaction  between  the  avatar and   User authentication credentials were then added
            its users. This integration was achieved by attaching   to  securely  link the  Unity  environment  to  the  OpenAI
            the OpenAI plugin directly to Unity, establishing the   account managing the API requests. This process required
            infrastructure for real-time conversational exchanges   the creation of a JSON file to store both the API key
            between the tutor and its audience. This phase required   and the organization ID, information provided through
            meticulous planning and execution to ensure that the   the OpenAI website. These credentials were critical in
            model’s sophisticated linguistic capabilities could be   facilitating interaction and ensuring that API usage could
            effectively translated into an interactive and immersive   be monitored effectively. Developers were mindful of
            educational experience.                            potential costs associated with API requests, as usage
              The first step in the integration process involved   is billed to the OpenAI account. This aspect required
            obtaining access to the OpenAI application programming   ongoing management to balance performance and budget
            interface (API) through the Unity engine. This connection   considerations.
            was essential for enabling the text-to-text interaction   A custom script was then developed in Unity to handle
            between the Da Vinci model and its users. A specialized   the text-based interactions. This script included a “using”
            package,  which  allows  for  the  incorporation  of  the   statement to access OpenAI’s API and a reference to the
            OpenAI API within Unity, was imported via the Unity   specific ChatGPT model employed for the project. The
            Package Manager using a Git URL. This unofficial package   script facilitated the transmission of user input – typed
            streamlined the process by providing a framework to   within Unity’s interface – to the ChatGPT model, which
            manage communication between Unity and the OpenAI   was configured to process and respond to the queries. For
            model.                                             this project, the GPT-3.5-turbo model was selected as it


            Volume 2 Issue 2 (2025)                         8                                doi: 10.36922/dp.8365
   53   54   55   56   57   58   59   60   61   62   63