Syllabus
We follow an applied-ML project through reading, building, data and evaluation, and writing. Each module asks how to do that work well with AI.
For each AI idea or practice we introduce, we also ask four questions: What problem was it solving? What is the proposed solution? Which older ideas help explain it? What is genuinely new? While tools and practices keep changing, this habit of thinking in “deltas” helps us reason about whatever comes next.
Learning objectives
The takeaways stack in order.
- how to use AI tools;
- where and why these tools work well;
- where and why these tools fail by construction;
- how to decide what to delegate, verify, redirect, or keep for oneself, based on where and why they work and fail.
Students should be able to do the following.
- check an AI-assisted reading against the source and build a map of the field;
- specify and decompose work for an AI agent;
- keep AI-co-authored work inspectable by preserving provenance and version history;
- manage an agent’s context and turn recurring workflows into reusable skills;
- form and test hypotheses about the mechanisms behind model failures;
- vet data, design evaluations, and audit the judge;
- write up a result as claim, evidence, and limits, and edit AI-drafted prose without losing authorship;
- justify with evidence their decisions to delegate, verify, redirect, or keep for themselves.
Schedule
| Lecture | Lab | ||
|---|---|---|---|
| Module 1: framing & reading | |||
| Sep 15 | Lec 1: what this course is | Sep 17 | Lab 1: close enough? |
| Sep 22 | Lec 2: reading matters | Sep 24 | Lab 2: connect four |
| Module 2: build & the harness | |||
| Sep 29 | Lec 3: vibe or spec | Oct 1 | Lab 3: project spec and reading map |
| Oct 6 | Lec 4: the abstraction ladder (of programming) | Oct 8 | Lab 4: fork in the road |
| Oct 20 | Lec 5: the harness | Oct 22 | Lab 5: giddy up |
| Oct 27 | Lec 6: why still build | Oct 29 | Lab 6: checkpoint |
| Module 3: data & evaluation | |||
| Nov 3 | Lec 7: outsourced data | Nov 5 | Lab 7 |
| Nov 10 | Lec 8: the machinery beyond vanilla transformers | Nov 12 | Lab 8 |
| Nov 17 | Lec 9: evals as experimental design | Nov 19 | Lab 9 |
| Nov 24 | Lec 10: auditing the artifact and the judge | Dec 3 | Lab 10 |
| Module 4: reflection & writing | |||
| Dec 1 | Lec 11: what the human had | Dec 10 | Lab 11 |
| Dec 8 | Lec 12: closing arguments | Dec 10 | Lab 12 |
- No lecture or lab the week of October 13, due to the Monday schedule shift.
- The last three labs run later than their lectures, due to the Thanksgiving holiday on November 26.
- The final two labs share the December 10 session.
- Topics may shift as the field moves.
Grading
- Grading: 30% lab checkoffs, 40% project, 30% personal portfolio.
- Lab checkoffs are similar to those in 6.390 and 6.101. Students typically check off in pairs, but each receives credit for their own work and explanation. Lab checkoffs are due in session.
- The project is graded on a one-page summary of the claim, evidence, and limits, plus the code repo. Existing UROP or MEng projects are fine as the course project. Students first confirm their advisor is okay with parts, the question, data, or code, being shared with staff and other students as the coursework requires. Students without a project can ask staff for sample ideas.
- The individual portfolio contains (1) three
SKILL.mdfiles, extracted from the student’s own workflow, and (2) three examples of the student’s own AI use that either support a claim the course made, or challenge it. - The project and the portfolio each have a checkpoint in Week 6 (week of October 26).
- AI use is assumed and not graded by amount. Work the student cannot justify earns no credit, however polished.