AAPT Summer Meeting 2026 PICUP Workshop

Using Large Language Models to Support Computational Integration

Welcome to the PICUP workshop on AI in computational physics!

This workshop explores how large language models and AI tools can support the integration of computation into physics education. Participants will discuss strategies for using AI tools to support instruction, curriculum development, and student learning in computational physics.

1:00-1:10 Introductions

Please sign in here
Become a verified educator - Use Standard Verification tab
Join Mattermost - email workshop coordinators for access
Join trinket.gopicup.org

1:10-1:30 Session 1a. An introduction to AI itself

Todd will discuss what AI is, what a large language model is and how it works, and how it can collaborate with you and your computer to write, run, and test code. Todd's slides

1:30-2:00 Session 1b. A demonstration of agentic programming

Walter will give a live demonstration of agentic programming, showing how a harness allows a human, a local computer, and a remote LLM can collaborate to do computational physics.

2:00-2:45 Session 2. Pondering the role of AI in your class

We’ll split into groups and discuss the various choices you will need to make about what role AI can play in your classes.

How should students be using AI tools to help them write code?
How should students be using AI tools to help them do physics?
How do you write a policy to tell the students how they should and shouldn’t interact with the robots?
How do you write a policy to tell the robots how they should and shouldn’t interact with the students?
How will assessment work?

2:45-3:00 BREAK

3:00-3:15 Session 3. The AI ecosystem

We’ll discuss what models are out there, what harnesses you can use to work with them, and the various ways to get access to this capability. What do you have to pay for, what’s free, and what systems will it run on? How did humanity get this capability, anyway?

3:15-4:45 Session 4. Doing an exercise with AI help

As our main working session, we’d like you to complete an exercise with the help of an AI, using the same rules and system prompt (if applicable) that you would have your class do.

A few choices, from simplest to most involved:

Orbital dynamics of two and three bodies -- As an example of a computational exercise that you might do in an introductory course and that is calibrated to be completed without that much LLM help, let's look at planetary orbits. This will be particularly interesting as a playground for how well system prompt guardrails work -- if you want AI to help students with programming but explicitly not computational science or physics.
Linear and nonlinear oscillators: the swinging pendulum -- This exercise begins with simple physics but that can quickly get into advanced topics. This exercise is one of my favorites; we're extending it a little bit given the fact that you all will be doing it with LLM help this time.
Nonlinearities in the vibrating string -- This might be sophomore or junior-level depending on how much help students get from their LLMs. This is a good stress test of your system prompts -- have you calibrated them so that the LLM will help the students with the things you want, but avoid doing things for them that you want them to learn on their own?
Dynamics of the Lennard-Jones gas -- This is a "graduate-level" exercise set without LLM help: mapping the phase diagram of the truncated Lennard-Jones gas. It will likely require substantial LLM assistance to code in an hour and a half, and it is probably a good demo of the physics you can learn without actually writing the code yourself.

4:45-5:00 Wrap Up

Organizers

Walter Freeman, Department of Physics, Syracuse University (contact: wafreema@syr.edu)
Todd Zimmerman, University of Wisconsin Stout (contact: zimmermant@uwstout.edu)

Partnership for Integration of Computation into Undergraduate Physics