CS372 Spring 2025, Assignment #1: Can LLMs Alone Lead to AGI?

Objective: This assignment invites you to critically evaluate a foundational question in AI:
Can large language models (LLMs) alone lead to artificial general intelligence (AGI)?

You are expected to examine arguments on both sides of the debate. In your response, treat “LLMs” as encompassing all systems derived from the language model itself, including prompt templates, instruction tuning, and tools that LLMs can directly call (e.g., calculators, code interpreters, search engines) through prompt-based control. However, components like external memory modules, sensorimotor interfaces, and symbolic planners should be considered outside the scope of LLMs.

Deliverables:

1. A two-part essay (1200–1500 words total):
   • Part A: Argument For LLMs Leading to AGI (600–750 words)
     Defend the position that LLMs, properly scaled and augmented through techniques internal to LLM training and prompting, are capable of achieving AGI.
   • Part B: Argument Against LLMs Alone Leading to AGI (600–750 words)
     Defend the position that LLMs are a necessary but insufficient component for AGI, and require integration with additional modules or architectures (e.g., memory systems, grounding via embodiment, planning modules).
   • Alternate Part B (600–750 words)
     As an alternative to Part B, you may argue that a completely new architecture is required for AGI, and that current LLMs are fundamentally limiting or counterproductive in this pursuit.


2. Cite at least three relevant papers, blog posts, or public talks (e.g., by Yann LeCun, Ilya Sutskever, David Chalmers, etc.) supporting your discussion.


3. You may reference chapters in the textbook such as SAGA, MACI, SocraSynth, Evince, CRIT, Dike vs. Eris, and Cocomo in support of either side of your argument.

Evaluation Criteria:

• Depth and clarity of reasoning (50%)
• Effective use of evidence and citations(20%)
• Engagement with technical and philosophical dimensions of the debate(20%)
• Quality of writing and organization(10%)

Submission Deadline: April 14th, 11:59pm

In-Class Debate: A live in-class debate will be held on April 14th to discuss key points and contrasting views from student submissions.