Syllabus for COMP560: Independent Research: Liberating chain-of-thought reasoning in large language models
Spring 2026
Dickinson College
Instructors: Matt Ferland, William Goble, John MacCormick
This syllabus is subject to substantial change up until the first day of classes. After the first day of classes, any changes will be clearly noted and announced.
Learning goals
Students will:
- be able to perform experiments on large language models;
- be able to document results of research experiments;
- be able to participate in team-based research;
- gain an appreciation of current research literature for large language models.
Fairness
Everyone in the course belongs equally to our research project community. The instructors aim to create an atmosphere where everyone feels a sense of belonging and feels free to ask questions.
Teaching methods
- Independent research based on guidance from instructors and other students.
- Regular ad hoc meetings of the full research team and smaller subteams as needed.
When and where
- There are no formal classroom sessions.
- A full team meeting will take place every Wednesday evening, 7pm-8pm. Attendance is recommended but not mandatory.
- Office hours: see the instructors’ office hour webpages: Goble, Ferland, MacCormick.
Books and resources
There is no textbook. Resources will be provided via the project website and a Microsoft Teams site.
Assessment and grading
The final grade will be assessed via four equally weighted marking periods (MP1-MP4):
- MP1 ends at 11:59pm, Tue Feb 10
- MP2 ends at 11:59pm, Tue Mar 3
- MP3 ends at 11:59pm, Tue Apr 7
- MP4 ends at 11:59pm, Fri May 1
For each marking period, a student will receive a score according to the following rubric. The notions of activity log, deliverable and research acievement are defined in more detail later.
| Score | Criteria |
|---|---|
| 90-100 | Activity log and deliverables provide convincing evidence of a minimum of 10 hours’ effort per week and include providing help to other students; contributions are of very high quality and at least some contributions represent research achievement |
| 80-89 | Activity log and deliverables provide convincing evidence of a minimum of 10 hours’ effort per week; contributions are of good quality but need not represent research achievement |
| 70-79 | Activity log and deliverables provide evidence of substantial effort (at least a few hours per week); contributions meet minimal expectations for quality |
| 60-69 | Activity log and deliverables provide evidence of substantial effort (at least a few hours per week); contributions do not meet minimal expectations for quality |
| <60 | Activity log and deliverables do not provide evidence of substantial effort |
An informal summary of these criteria is as follows. You can get a score in the B range purely by devoting a reasonable amount of effort to the project. In this range, effort is far more important than achievement. To get into the A range, you will need to help other students with the project and produce at least some outputs that go beyond pre-existing results. Strikingly original research is not required, just something that demonstrates an ability to try something new and analyze the results – something that can be considered an achievement.
Activity log
An activity log is a private Microsoft Teams channel in which a student keeps a log of all activity for the course, preferably updated several times per week. Activity logs are discussed in more detail on the separate activity log page.
Deliverables
A deliverable is any output produced by work on the project. Deliverables include:
- source code checked in to a project repository;
- experimental results checked in to a project repository;
- documentation, discussion, analysis, and other technical documents checked in to a project repository (edits to existing documents also count);
Research achievement
A research achievement is any deliverable that goes beyond existing scientific results or techniques. Typically a research achievement demonstrates some original thinking. For example, research achievements could include: a new type of experiment; a new explanation or analysis of an existing experiment; or a code feature that adds some new scientifically meaningful functionality. A research achievement need not be strikingly original. It can closely resemble existing work, but must demonstrate some extension of pre-existing content.
Helping other students
It is an expectation that experienced students will provide substantial amounts of help to inexperienced students. Helping others is a highly valued activity and should be emphasized in a student’s activity log. To achieve an outstanding grade in the course, it will be necessary to provide substantial assistance to other students.
Grade threshold
The following thresholds, or possibly more generous thresholds, will be used for final grades: 93%=A; 90%=A−; 87%=B+; 83%=B; …; 60%=D−.
Plagiarism, copying, collaborating, and AI
Use of all relevant AI tools is encouraged and expected for this research project. AI use should usually be acknowledged, although there are exceptions such as boilerplate code.
All work can be done in teams or individually. Teams can change on an ad hoc basis during the semester. As with any other scientific research, joint work should be attributed appropriately to all contributors.
The College’s standard policies on plagiarism apply, and you should be familiar with them.
Accommodations
The instructors will follow college policy on Accommodating Students with Disabilities.