Meeting Summary: CS 486-686 Lecture/Lab, Spring 2025

Date: January 30, 2025, 08:06 AM Pacific Time (US and Canada)
Meeting ID: 893 0161 6954

Quick Recap

Project Evolution:
Greg announced the evolution of the Hello World Chat Assistant project, emphasizing the importance of domain expertise in coding.
AI in Research & Chat Assistants:
He highlighted both the challenges and potential benefits of using AI tools for understanding research papers and for creating chat assistants.
Python Overview:
An overview of the Python programming language was provided, including its features and its role within the Gen. AI landscape.

Research Paper:
Begin reading the assigned research paper on chain-of-thought prompting.
Chat Assistant Projects:
Start working on individual chat assistant projects.
Creative Brainstorming:
Think creatively about unique features or real-world applications for the chat assistant projects.

Python Resources:
Provide Python resources for students to review or learn from.
Demonstration:
Demonstrate light LLM multimodal capabilities on Thursday.
GitHub Repository:
Create a GitHub repository for students to submit paper evaluations and approve access requests for Render deployments.
Evaluation Template:
Offer a Markdown template for paper evaluations.
Focused Instruction:
Dedicate Tuesday’s class to a more in-depth discussion on Python and related technologies (FastAPI, LightLLM).
Lab Support:
Collaborate with Steve to provide hands-on help during next week’s lab section.

Project Weight and Deadline:
The project has become a more significant component of the coursework, now with a higher grading weight and a two-week deadline instead of one week.
Engagement and Learning Curve:
Since the first lab session was less engaging than anticipated, additional time will be provided for students to familiarize themselves with Python. Resources will be offered for those who need to refresh their Python skills.
Web Interface Considerations:
Options for building a web interface were discussed, including decisions on whether to stream responses or deliver a single response.
LLM Usage:
The plan is to use large language models for the project side, emphasizing the balance between leveraging models and developing domain expertise.

Reading and Critical Evaluation:
The importance of critically reading research papers was underscored. Greg introduced the idea of using AI tools to assist with understanding research content while maintaining a healthy skepticism.
Repository for Evaluation:
A repository will be created for paper evaluations, accompanied by instructions for its use.
Learning Opportunity:
Students are encouraged to use AI tools as a way to develop critical thinking skills, particularly when verifying and questioning AI-generated content.

Tool Experiences and Concerns:
Experiences with AI platforms such as Google Gemini and OpenAI were shared, including concerns about the terms of service, especially regarding data usage.
Shared Account Considerations:
The potential for creating a shared account was mentioned, though with awareness of possible violations of service terms.
Google Gemini Demonstration:
A demonstration showcased Google Gemini’s capacity to generate detailed, referenced documents.
Render & GitHub Integration:
Issues with integrating Render with GitHub were noted, and future deployment attempts are planned.

Interpreted vs. Compiled Languages:
Discussion compared Python’s interpreted nature—compiling to bytecode at runtime—with the direct binary compilation of languages such as C and C++.
Global Interpreter Lock (GIL):
The existence of the GIL in Python, which restricts the simultaneous execution of multiple threads, was explained. It was noted that performance can be enhanced using libraries like NumPy and TensorFlow.
Python 3.13 Update:
A recent update in Python 3.13 has removed the GIL, enabling truly parallel threaded execution.
Further Exploration:
Plans were made to explore Python’s applications further, particularly in the context of the Gen. AI landscape.

Vercel Deployment:
The use of Vercel for deploying web applications was covered, including necessary approvals and the configuration process.

Language Features:
Greg reviewed the origins and features of Python, noting that it is dynamically typed and emphasizes a simple syntax centered around data structures, functions, and classes.
Decorators and Data Types:
The importance of decorators in web frameworks and a solid understanding of built-in data types like lists and dictionaries was highlighted.
Industry Relevance:
Python’s widespread use in industry was emphasized, along with the recommendation for continued practice and skill development.

Effective Prompt Construction:
The need to build effective prompts for large language models (LLMs) was detailed, with a focus on achieving desired responses.
Chat Assistants and Automation:
Chat assistants were identified as a key component in programming with LLMs and were discussed in the context of potential automation.
Additional Concepts:
The topics of embeddings, vector databases, autonomous agents, and model fine-tuning for specific tasks were introduced.
Industry Transformation:
The growing importance of chat assistants in transforming customer service and other industries was noted.

Incremental Development:
It was suggested that the chat assistant should be built incrementally to mitigate potential issues.
Innovative Features:
Brainstorming ideas for unique functionalities was encouraged, including a side-by-side comparison feature that could display responses from different models simultaneously.
Individual and Group Projects:
While individual work is required initially to build hands-on experience with APIs and technologies, there was mention of a potential group project for final presentations.
Project Kickoff:
Students are encouraged to begin reading the research paper and to start working on their chat assistant projects immediately.