Understanding recent AI systems

Part III, AI 100, 2026W1

Learning outcomes

Students will be able to:

1. Describe the training and inference processes of large language models.
2. Explain how alignment modifies base models to create chatbots.
3. Summarize how diffusion models generate images from noise.
4. Differentiate reasoning models from general text-generation models.
5. Explain current AI applications in robotics and their limitations.
6. Evaluate the effectiveness and limitations of alignment strategies for AI systems.

Theme: Learn how state-of-the-art AI models work, including language, vision, and robotics systems, and how they are aligned to human needs.

1. Identify and explain common sources of hallucinations and reasoning flaws.
2. Define the “alignment tax” and illustrate examples of safety-performance trade-offs.
3. Assess the reliability of AI outputs in specific application scenarios.

Theme: Identify and analyze the technical and practical limitations that challenge the reliability and safety of modern AI.

Select topics

• How do large-scale, pretrained generative models work?
• Applications of generative models:
• How do chat models work?
• How do image generation systems work?
• How do reasoning models work?
• How are AI systems aligned to human preferences?
• What is the current state of the art in robotics?