Concepts

We introduce some important terms used across the entire system before detaily introducing how the system is implemented.

Course, Chapter, Module

That are various terms that refers to a teaching process. For consistency, we conclude the scope of lecture in three categories:

1. Course. Which refers to the complete set of lectures that often spans across an entire semaster.

2. Chapter. Which refers to the largest divided chunk of course.

3. Module. Which is used to refer to all divisions of chapters and modules.

For example, A course might have the following structure:

• Towards Artificial General Intelligence

  • What is AGI?

  • Trends in LLM

    • Prior to LLM, what is PLM?

      • BERT

      • BART

    • Pretraining

      • GPT3

    • Intruction Tuning

    • RLHF

  • Trends in VLM

  • etc.

This translates to the following correspondence:

Course	Chapter	Module	Module
Towards Artificial General Intelligence	What is AGI?
Towards Artificial General Intelligence	Trends in LLM	Prior to LLM, what is PLM?	BERT
Towards Artificial General Intelligence	Trends in LLM	Prior to LLM, what is PLM?	BART
Towards Artificial General Intelligence	Trends in LLM	Pretraining	GPT3
Towards Artificial General Intelligence	Trends in LLM	Intruction Tuning
Towards Artificial General Intelligence	Trends in LLM	RLHF
Towards Artificial General Intelligence	Trends in VLM

Lecture

We found common that different teachers reveals different preferences in how fine-grained should their course be divided. For example, Prof. Liu for TAGI tend to provide longer lectures (Forming all the contents within the Trends in LLM into one lecture) to ensure the related concepts are introduced at once, but for Prof. Zhan prefers to further divide the subject so the students can quickly navigate the contents for HSU (How to Study in University).

To ensure our system could support such diverse demands, MAIC do so via independing the concept of lectures with course/chapter/modules. We use the concept of Lecture to refer to a consistent teaching process. This ease off the complexity for algorithm design and implementation. Instead of worrying the human teacher would change the structure of the course at any time and frequently causing the demand for the system to adapt to it, the algorithm researchers can now intuitively assume such revisions in teaching plan mainly occures in a consistent lecture.

Agenda

In discussing the creation of an online educational system, it’s crucial to consider how we facilitate communication between teachers and students, especially in terms of conveying the teachers’ intentions effectively. Imagine a physical classroom setting where a teacher meticulously prepares a wealth of materials and consolidates them into a teaching plan. This plan serves not just as a content with rich knowledge but as a structured learning trajectory that the teacher aims to guide the student throughout a lecture. The interactions, which students poses during lecture, might vary, but the teaching plan remains constant, serving as the general knowledge to regulate the lecture.

With this in mind, we’ve taken inspiration from this traditional approach to craft the core controlling info. We call this the Agenda. Just as a physical teaching plan helps structure a realworld lecture, the Agenda serves to encapsulate the teaching intentions in a structured format online. The reason for structuring it as a tree with detailed planning at each node is to ensure that the educational goals are clear and systematically organized. This not only aids in the delivery of the content but also simplifies future revisions and updates to the teaching plan, as adjustments can be made efficiently and coherently.

We chose the term “Agenda” because it conveys a sense of planned and organized activity, much like the agendas used in formal meetings and events. In the context of education, the Agenda lays out a structured plan that guides the teaching and learning process, ensuring that the educational objectives are met effectively. This structure helps to bridge the gap between the algorithms at the teacher’s end and the students’ end, ensuring that the instructional flow is smooth and coherent. Thus, the Agenda is not just a tool but a vital conduit through which teaching intentions are clearly and effectively transmitted to students, making the learning process more organized and impactful.

In practice, by employing DFS (Depth First Search) to the agenda, we can retrieve a learning trajectory planned by the teacher. To cut it short,

\[ \text{An Agenda is a teaching plan organized as a tree.} \]

Where,

1. \(\text{The nodes in the DFS order is the learning trajectory;}\)

2. \(\text{The parent node concludes the knowledge of its childrens as an abstract.}\)

Function and Action

The design of agenda captures the teaching plan from a learning trajectory level, but different students may react to the same trajectory nodes differently. To provide an interemediate controlling unit between the hyper-leveled concepts in the learning trajectory and detailed interactions, we introduce the concept of Function and Action.

\[ \text{A Function is a small set of actions in class that contributes to a teaching activity} \]

\[ \text{An Action is an atomic interaction occured in the lecture process} \]

For example, a classic function in MAIC is the \(ReadScript\) function. Image where the teachers makes a statement/introduction by reading a pre-written script, where the student may chose to do nothing or make responses by saying something, and other agents may then intereact with the students. In such process, each time a character (including the student user) said something (including the time which the teacher read the script) is formalized as a speak action. These interactions, conducted by the characters and centered around the script, togetherly contributes as a single \(ReadScript\) instance organized as \(ReadScript(script)\).