Grading & Feedback Fine-Tuning Launching March 26th!

“I think the AI industry has trained users to treat interactions with AI as an iterative conversation, where results get refined as we go… I would benefit from the ability to iterate a bit more and refine the results as I go.” - Adam Kleinbaum, Dartmouth College

That’s the quote I received last term when collecting user feedback from our pilot instructors. It makes a lot of sense thinking about just how many iterations I need to go through to get ChatGPT to give me something “good”. Now I’ve trained my brain to send in something totally vague and refine the results further and further.

With TimelyGrader, we might have landed on the wrong side of efficiency and become more of a 'one and done’ type of thing. While we encourage iterations to ensure accurate grading and feedback, the user experience isn’t as seamless as it could be. There are a lot of customization options for grading and feedback and that makes it more time-consuming to be iterative.

Here’s how we are supporting iterations: grading and feedback fine-tuning

1. Instead of immediately uploading student submissions once assignments are created, there will now be an option to fine-tune the assignment where instructors will be asked to upload up to three student submissions. Of course, they can also skip this entirely.

2. Once processed, instructors will be able to view first-pass grading suggestions and feedback. The first generation is never perfect and instructors are bound to find areas that the AI missed because of hallucinations, missing context, or sometimes due to a vague rubric. While in fine-tuning mode, instructors can make edits immediately.

3. Say the rubric is vague and could use a bit more detail or there was a mistake made on the rubric and the instructor wants to change it, they can simply ask the AI to enhance the rubric based on desired changes. For example, they want to quantify each rating so the rubric is more objective. Additionally….👇👇👇

4. They can also add additional grading instructions and feedback instructions directly during the fine-tuning mode to provide specific context to the AI. Once satisfied with their changes, instructors can re-generate grades and feedback to see if their changes made any impact.

   

5. Rinse and repeat until grading and feedback is accurate enough and move onto grading!

   

That’s it for our final preview for grading and feedback fine-tuning, coming to all free, paid, and institutional users! You will be able to fine-tune to your heart’s desire this March 26th, 2025 at 5PM PST.

PSA: Important for all current and prospective users.

We are extending our legacy account upgrade for another week. If you want to join our 300+ users who received their free legacy account upgrade, please fill out this quick survey: https://forms.gle/zjUkoxFRfPrznfah9.

If you have already requested an upgrade, a confirmation email should have been sent out earlier this week. If you did not get a confirmation, please submit another response and make sure your email address is correct!