Abstract:

When sophisticated prompt engineering techniques are combined with a structured data engineering approach, OpenAI’s ChatGPT4 generated content will meet and beat human master levels in term of relevance, quality and accuracy. This elevates the AI-assistant role in adult learning from learning management and scheduling to a completely new level of AI mentorship.

We apply AI, and ChatGPT4 in specific, to achieve the optimised mix of the following three goals:
(i) increase efficiency by accelerating learning progress;
(ii) increase effectiveness in terms of gaining comprehensive and in-depth knowledge of the study subject;
(iii) furthermore design a learning curve customized to the specific needs of the mentored human, that maximises the content coverage for the time available and hence optimize exam preparedness.

After building a prototype, based on a successful PhD-level course and the prompt engineering journey, we reinvest our experience in building a general purpose AI-mentor model that accelerates the learning curve and enhances the learning experience for a broader range of courses and students.

ChatGPT4 features include the interpretation of file attachments, and mathematical formulae uploaded in image format in particular. GPT Builder and custom instructions enable the creation of the customized 'AI Mentor' focusing on a specific content. Prompting techniques, including role-based prompting with reference to the RICCE framework (i.e. [Role], [Instructions], [Content], [Constraints], [Example]) and Q&A sessions, were used to identify and address unclear areas in the body of knowledge where recap was required before progress could be made.

In addition to zero-shot and few-shot prompting, advanced prompting techniques, with reference to research articles were involved and proved successful in testing and ascertaining reasoning capabilities of ChatGPT4, and further enhanced when implementing Chain-of-Though (CoT) logic introduced in Wei et al. (2022), Tree-ofThought (ToT) techniques proposed by Yao et el. (2023) and Long (2023). and Generative Knowledge Prompting by Liu et al. 2022.

The technical viability leverages the advanced features and plugins of ChatGPT4 with a “Prompt Engineering Mindset” to generate the highest output quality. A structured approach by the AI mentor is taken to build the knowledge profile of the human to detect and bridge the knowledge gaps from the targeted knowledge architecture. Progress and results are measured using SMART KPIs and a performance measurement model in MATLAB®.

As regards actual or direct outcome of the approach, an AI-mentor (AI tool) can be developed that facilitates learning of a variety of subjects in a more efficient and more effective way, customized to the particular needs of the learning environment and subject.

As conclusion, our concept discussed in this paper is unique, fundamentally different from AI-assisted frameworks, and has the potential to “hack” or revolutionize self-directed learning in a custom-fit and ethical way. Our approach can be directly integrated into knowledge-based university training programs, vocational and professional training and certification programs and potentially extended to other levels of education.

Keywords:

ChatGPT4, AI Mentor, prompt engineering.