Artificial General Intelligence

This course explores theories and methodologies for developing AI systems capable of human-level general problem-solving abilities. Students will study various approaches including large models, program synthesis, meta-learning, representation learning, and neuro-symbolic methods, while examining current research trends in improving abstraction and reasoning capabilities. The course includes practical projects where students design and implement general-purpose AI systems.

  • Time: Mon/Wed 1:00pm-2:30pm (Fall 2026)
  • Venue: TED-Hall
  • Codes: AI5102
  • Prerequisites: Students are expected to have solid understanding and hands-on experience on machine learning and deep learning (AI5213, AI5214).

Notice

  • The semester will begin at 8/31. Enjoy your vacation!
  • Access the AI5102 Ed Discussion forum. If you haven’t already been added to the class, use this invitation link.
    • Use your full name in the format “Firstname Lastname” with the first letter of each name capitalized (e.g., Sundong Kim).
    • Don’t forget to use your valid GIST email address.
  • Submit your assignments at the AI5102 Gradescope. Entry code is 4DBD25. Make sure to enter your correct student ID.
    • Use the same name format as for Ed Discussion: “Firstname Lastname” with the first letter of each name capitalized. (e.g., Sundong Kim).
    • Make sure to enter your correct student ID and your valid GIST email address.

Syllabus

We don’t have official textbook, but we will cover papers relevant to the topic. Main readings will be based on recent research papers and technical reports in the below syllabus. For some major topics, following the staff’s opening lecture, students will present papers covering recent trends, allowing for in-depth discussion of the subject.

Week Description Readings Homeworks
8/31      
9/2 Introduction to AGI and Current Landscape   Sign-up Survey
9/7 Game Intelligence as a Path to AGI: Othello as a Testbed    
9/9 Game Intelligence as a Path to AGI: Othello as a Testbed  
9/14 GI as a Path to AGI: Othello as a Testbed: Detail MuZero  
9/16 The Five Tribes of Machine Learning Tribes  
9/21 A Minimalist Guide to Program Synthesis    
9/23 Neuro-Symbolic Program Synthesis Paper
9/28 On the Measure of Intelligence Paper  
9/30 Building Sys-2 Reasoning Capabilities in Three Steps    
10/5 Holiday    
10/7 Model-Based Reinforcement Learning Blog  
10/12 Mastering Diverse Domains through World Models Paper  
10/14 Model-Based RL: How do humans learn so fast and efficiently?    
10/19 Midterm Week (No lecture)  
🏁 🏁      
10/26 Model-Based Reinforcement Learning: Wrap-Up    
10/28 LLM for Mathematics Paper  
11/2 LLM with planning Paper  
11/4 LLM RLHF Paper  
11/9 In-Class AGI Write-up (HW3 Alternative)    
11/11 LLM Parameter finetuning Paper  
11/16 LLM Interpretability Paper  
11/18 LLM Alignment Paper  
11/23 Invited Talk: Brain-Inspired AI NeurIPS-23 ICLR-25  
11/25 Continual learning Paper  
11/30 Continual learning Paper  
12/2 HW3 review & discussion    
12/7 HW2 review & discussion  
12/9 Invited Talk: RL    
12/14 Final Week (No lecture)    
🏁 🏁      

Staffs

Communication: The course schedule and all resources will be posted on this course website. All class discussions, announcements and other communication will take place via Ed Discussion. Use the public comments if your question is relevant to the course material. We aim to respond to questions within 2 business days, often sooner. If you need to contact the course staff, please make a private question on Ed instead of sending a personal email. You’re free to visit TAs during office hours.

Sundong Kim

Sundong Kim

Instructor

QnA: After class (Lecture room) or Ed discussion
Office: AI Building (S7), Room 204

Homeworks

Submit your assignments at the AI6105 Gradescope.

Gradings

You will earn A if (but not only if) your score is at least \(80\times(1-ε_1)\%\) of the total score, B if your score is at least \(60\times(1-ε_2)\%\), C if your score is at least \(40\times(1-ε_3)\%\), for some \(ε_i ≥ 0\) to be determined later. All participants in the course are evaluated equally, regardless of the course codes.

GIST-logo