AI Policy

AI Policy

or: How to Waste $40,000+ by
Asking ChatGPT to Do Your Homework

Summary of My Policy

  • Generative AI (including, but not limited to, ChatGPT, Claude, and Copilot) will not be permitted for any submitted work, in any capacity.

Summary of My Policy

  • If a student is suspected of having used generative AI to assist with coursework, in whole or in part, any impacted work will either be given a zero or replaced with a closed-notes, proctored, written assignment at my discretion.

But Why?

But Why?

  • Easy answer: I'm not interested in whether ChatGPT can solve problems (it can*)
  • For you to be an effective computer scientist, you need to understand the tools yourself

*depending on your meaning of "solve"

But Why?

  • Complicated answer: the plagiarism machine is only capable of plagiarism
  • "the purpose of a system is what it does"

But Why?

  • Even more complicated answer: it is making you measurably worse at computer science!

"Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task" [https://arxiv.org/abs/2506.08872]

  • Kosmyna, N. et al., Jun 2025
  • A few highlights from the abstract:
    • "LLM users [...] struggled to accurately quote their own work."
    • "Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels."
    • "EEG revealed significant differences in brain connectivity: [...] LLM users displayed the weakest connectivity."

Perception Is Not Necessarily Reality

"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity" [https://arxiv.org/abs/2507.09089]

  • Becker, J. et al., Jul 2025
  • Primary result: "Before starting tasks, developers forecast that allowing AI will reduce completion time by 24%. After completing the study, developers estimate that allowing AI reduced completion time by 20%. Surprisingly, we find that allowing AI actually increases completion time by 19%--AI tooling slowed developers down."

Why Should You Care

  • When you leave college, your grades won't matter, but your ability to apply the skills you learned will
  • If you rely on LLMs to shortcut the learning process, you won't develop those skills yourself

Why Should You Care

  • LLMs are plausible text generators. They organize words and characters into a response that fits the prompt in a purely mathematical sense.
  • Keep in mind the most important fact: When LLMs are correct about something, that is incidental, not by-design.

Part 2

Unproductive Dunking

twitch.tv/gpt_plays_pokemon

  • A ChatGPT instance trained specifically on Pokemon Red, one of the most well-documented games on the internet
  • ChatGPT had information about internal mechanics, maps, and more that the casual player wouldn't have
  • It took ChatGPT more than 80 hours to earn the second badge in the game. Average human time to complete: ~3 hours
OpenAI's "smartest model to date"

ChatGPT vs the Atari 2600

How Many "B"s Are in "Blueberry"?

  • The answer is: three
  • Remember that the goal of ChatGPT is to give you feedback that has the same shape as a correct answer, in a linguistic sense.
  • ChatGPT does not reason about things the way that you or I do, and therefore cannot be trusted to do more than copy & paste the same answers you would get from your own Google search!