FeedBot

What is FeedBot?

FeedBot is an automated feedback tool that we are using this semester. It uses a Large Language Model (LLM) to provide small suggestions based on your responses to problems of an assignment (when enabled: not all problems may get feedback).

This feedback will be provided via a link on the Gradescope assignment page. You should find a link under a rubric item titled “FeedBot(0/0)”. Note that this does not have any bearing on grading (hence the 0/0), and is intended to be used on in-progress code: so submit once you have something you think is good. It should look something like this.

At the link, for each problem, you will find your response for that problem along with a few lines of feedback that you may use as a suggestion to make changes to your submission.

Sample output

Consider the following problem from Lab 4:

Here is a response for an incorrect/incomplete submission:

Note that there are several problems: we have no purpose statement and no tests, but FeedBot is designed to always focus on the first issue in the Design Recipe that needs help, as that should help you to correct issues in the best order, and won’t comment on issues later in the Design Recipe. Once we fix this issue, we can get more feedback:

However! You are limited to only 5 responses from FeedBot (and there is a 1hr delay between them), so please do not try to use this to do your HW for you! Do as much as you can on your own.

When to use this

We hope you use this to help with assignments that you have worked on, but haven’t completed yet. It’s not intended merely as a “final check” to ensure adherence to the Design Recipe, though you can use it that way.

In any case, you will be restricted to 5 FeedBot results per assignment, so we’d suggest you do significant work before you submit for the first time.

You may still submit your assignment more than 5 times, but you will only receive FeedBot results for 5 of the submissions. In order to not “waste” responses on minor fixes, there is a 1 hour cooldown period after each submission until the next submission that can get a FeedBot response.

How should you use this feedback?

FeedBot suggestions are just suggestions. You do not need to strictly follow its advice, and your assignments will still be graded manually and/or with auto-graded tests with no consideration for FeedBot results. The feedback is only to help you, and we hope that most of time, FeedBot will provide useful suggestions based on your adherence to the function and data design recipes. That being said, getting a “Well done, looks good.” does not equate to full points for that problem, and conversely, you may still get full credit for a problem for which FeedBot asks you to change something.

Use your best judgement when following FeedBot advice. Sometimes, LLMs may “hallucinate” or provide unhelpful or potentially harmful feedback. It is your responsibility to recognize this. You should ignore FeedBot suggestions if you think they are misleading.

Please DO NOT submit regrade requests with the reason “FeedBot told me to do XYZ”. If you believe that the FeedBot advice is misleading/unhelpful for any particular question, please let us know (we’ll try to refine the tool!). Also – use the buttons with the feedback to give us direct feedback on individual pieces of advice. If you are unsure whether to follow the feedback, consult a TA/professor for guidance.

Who has access to the feedback?

FeedBot results for a submission can only be accessed by the person who submitted to the assignment, TAs and professors. You may share the link to your results to any TA to clarify any doubts. Do not show your results to anyone else in the class, since it contains your submission code.