Assignments regraded in Brightspace
Panos Ipeirotis had Dot go through an assignment a teaching assistant had already graded. Dot was to find what the assistant had missed, check AI answers with Claude, and change the grade and comment in Brightspace.
Done with DotAdvanced
What was done
On 3 October 2026 Panos Ipeirotis writes that the teaching assistant had already graded the assignment, and had done a good job. The feedback was scarce and limited to why points were deducted. He still asked OpenAI's Dot to regrade everything, find what the assistant had missed, verify AI-generated answers with Claude, and then revise grades and feedback directly in Brightspace. The profile says New York.
How it went
He writes that this is the first time he trusts the setup enough to imagine that grading will not be a task for the assistants going forward. The post does not say how many submissions were changed, or by how many points.
Worth knowing
The grades were changed in the learning system. The post does not say a colleague read the changes. A student's grade should be explainable by a person.
How to do it
- Ask the agent to propose changes beside the existing grading, not to write them in at once.
- Read the difference. Ask it to show what it checked with another model.
- Enter the grade yourself, or approve each difference before it writes.
Try it with Dot
Read the submission in [system] and the existing grading. Propose what you think is missing. Do not change the grade. I will read each difference before anything is written.
Replace text in brackets with your own.
Done when. Each proposed change has been read by you before it stands as a grade.
Safety rule. The agent does not change the grade until you have approved the difference.
Read the original
Read the original · Panos Ipeirotis on X ·