Using LLMs on the Final

For clarity, I use all-caps MAY, MAY NOT, and MUST below, just like in the RFCs.

The rules

• You MAY (but are not required to) consult LLMs for any portion of the final exam.
• You MAY NOT allow LLM-based agents to take any action on your behalf other than producing text or images in response to your prompts.

• If any part of your exam submission comes from or is substantially inspired by an interaction with an LLM, you MUST include an appendix consisting of a list of entries, each of which contains:

  • what portion of your exam are we talking about?
  • what prompt(s) did you use to generate the relevant portion of the exam?
  • which LLM(s) did you consult?

Examples

Here are a few examples of applications of these rules. This is not a comprehensive list of possible uses of LLMs on this exam, but I hope it helps clarify the rules.

• You MAY ask an LLM to explain something.
• You MAY ask an LLM to write some code.
• You MAY ask an LLM to draw a diagram.
• You MAY ask an LLM to suggest some command-line syntax.
• For all of the above examples, if you include the LLM's output in your exam (even if you edit or expand upon it), you MUST include a corresponding entry in your appendix.
• You MAY NOT ask an LLM to login to the Fake Discussion Forum, submit a post, or extract cookies from your browser.
• You MAY NOT ask an LLM to execute an openssl command on your machine (or any other), even if it has provided you with the command-line syntax. You'll need to execute that baby yourself.

Some general principles to consider

• The goal is learning.
• LLMs can help point you in the right direction, but if they do all the work for you, they inhibit your learning.
• LLMs often produce wrong or incomplete answers. You need to check up on them via additional research and/or experimentation.
• You are responsible for the content of your exam submission.