Overview¶
In this class, you grade your own homework assignments. After each submission deadline, a key is posted, and then you compare what you submitted to the key. Based on that comparison, you submit to the professor what you believe is the appropriate grade, along with a corrections document that describes what you got incorrect and including the correction.
Motivations¶
LLMs¶
One motivation for self-grading is ChatGPT and other Large Language Models (LLMs). The reality is that these generative AI tools have gotten so smart that they can, properly prompted, correctly solve virtually any question on any assignment that you will get in this class. That includes both conceptual questions and coding-based ones. Since completing the HW assignments correctly and on time contributes to your semester grade, this entices many students to simply copy-paste the HW prompts into the LLMs, then copy-paste back in their answers. This results in full completion of the assignment and zero actual learning.
Having you grade your own submitted assignments and to submit corrections forces you to revisit what it is that you submitted (whether or not it came from your brain in the first place), compare it to the correct solution, and to the extent that it is incorrect, reflect on where your answer went astray (again, whether or not you actually generated it).
To be clear, even this is not LLM-proof: it’s possible to paste in these instructions, the key, and your submission, and have the LLM do the grading and corrections for you well. In that case, all I can say is: I feel sorry for you.
Instructor time and timely feedback¶
Two other motivations for self-grading are instructor time and timely feedback. Alas, your professor does not have enough time to properly grade each student’s homework assignment in a timely fashion. Self grading saves your professor time and also avoids the scenario where you’re waiting weeks to get grades back.
Spaced repetition¶
Another motivation for self-grading is spaced repetition. A well-established result from cognitive science is that learning a topic goes better when you review the material at spaced intervals in time, rather than just seeing it once followed by a massive cram of all the topics just before an exam. By having HW due dates one week after the class session on that material, followed by the self-grading and corrections due a week later, you engage with the material more in total and over a longer duration than you would otherwise.
Instructions¶
Grading¶
For each question, compare the answer you submitted to what the answer key provides. In addition to the correct answers, the answer keys provide point values for each prompt, by treating each distinct element of each problem as worth one or more point. For each of item that is worth nonzero points, evaluate fairly whether you got it correct (full points), incorrect or didn’t do it (0 points), or substantially but not completely correct (half the points). Sum these up, divide by the total possible, and that’s your percentage score.
For open-ended questions, do not get bogged down in the details. It’s fine if you didn’t format your plots the exact same way, or if your interpretation is different than the one provided by the key. Just use reasonable judgment of whether your approach was reasonable, scientifically and mathematically justified, and articulated/displayed sufficiently well.
In all of this, the watchword (watch-phrase?) is good-faith. Make a good-faith attempt at the assignment in the first place, followed by a good-faith evaluation of what you got right and what you did not.
All assignments carry the same weight for your grade, irrespective of their number of raw points. As such, it’s only meaningful to compare percentage scores, not raw point totals, across assignments.
Corrections¶
For each element that you did not get 100% correct, you will write an explanation of what you did that was incorrect. If the prompt involves a calculation, you will replicate the calculation yourself that the answer key provides---even if that’s just copy-pasting the code from the key into your own Jupyter or Colab session and running it.
For each assignment, compile all these corrections into a standalone document. You will submit this along with your self-reported score.
Submission¶
For each HW assignment, when the answer key gets posted, there will also be posted a link for where you submit your self-graded score along with your corrections. The due date for submitting these scores and corrections will be posted at the same time.