Format and core idea

The seminar runs as Hausarbeit mit Referat – a written essay plus a presentation – but the format departs from the traditional seminar in several ways. The central assumption: students will use LLMs throughout their work. The seminar does not try to detect or prohibit this use. It teaches and assesses the judgment needed to use these tools well, and it places the graded moments where that judgment is visible.

In class, we develop taste: the ability to recognize good writing about security and privacy and distinguish it from bad – through editing exercises, close reading, and live critique. Taste is the didactic orientation of the seminar, but not the grading criterion. Grading happens against a structured assessment scheme with specific dimensions (see Assessment below).

Any tools may be used at any stage of the work. The bar is the quality of the final output and the editorial judgment visible across the three assessment components. LLM-generated text without editorial engagement will typically not meet the requirements of the assessment scheme.

The deliverable

Participants write a 2500-word essay explaining a non-trivial security or privacy concept clearly and rigorously for an advanced undergraduate audience. Topics are curated by the instructor, announced in week 1, and assigned in week 2 after participants rank their preferences.

Workflow over the semester

Week 9 in detail

Before the session. The first-pass draft (at least 1500 words) is submitted at the start of the session. Timely submission of the first-pass draft is a prerequisite for the remaining assessed components, since neither the pair editing nor the supervised revision session in week 14 can take place without it.

During the session. Participants are paired for the pair editing exercise. Pair editing uses symmetric pairs: both partners work on each other’s drafts mutually, in the same room, with the instructor circulating. This pairing is independent of the reader-writer assignment for the Referate – pair-edit partners and Referat readers are deliberately different people. The pair edit is a formative, low-stakes workshop pass; the Referat is a summative, public critique.

After the session. The pair-edit feedback is integrated and a revised draft submitted by end of day. This version is the one the assigned reader will work with for the Referat. LLM use during this revision is allowed and expected – the week 14 supervised session is the one that must be clean.

After submission. Reader-writer assignments are announced at the end of week 9, after all revised drafts are in. Until that point, nobody knows who will be reading their draft.

The three assessment components

The seminar assesses three distinct skills.

1. Explanatory writing (draft and final essay)

The quality of the submitted text across its revisions: explaining a security or privacy concept clearly, rigorously, and economically.

2. Substantive reviewing (the Referat)

The ability to read someone else’s work well. In the Referat, participants present a critical reading of another participant’s draft – not their own work. The Referat tests the ability to read a technical text closely, diagnose its strengths and weaknesses, and discuss them publicly – not primarily topic research.

The reader chooses one of three modes for the Referat, depending on what the assigned draft actually needs:

Choosing the wrong mode – faithful explanation of a bad draft, or critique of a good one – is itself a sign of weak judgment and will be reflected in the grade.

At the end of each Referat, the instructor asks the reader to improve one specific sentence or paragraph from the writer’s draft on the spot: 90 seconds of think time, no devices, then the improved version delivered aloud.

3. Revision under critique (supervised revision session, week 14)

The ability to engage substantively with criticism of one’s own work – demonstrable engagement with critique, not just text quality.

The week 14 session runs for 180 minutes in the seminar room. At the start, participants receive three things:

• Their own draft
• Their notes from the Referat on their work (captured on instructor-provided paper during weeks 12–13 and held by the instructor until week 14)
• A reviewer report from the instructor identifying exactly two substantive issues with the draft: one clarification or structural point, and one substantive challenge that requires real thought

The session is used to produce a final version and a short response letter addressing the reviewer’s two issues and the peer feedback. No internet, no LLM access. The instructor circulates and answers practical questions but not content questions.

The reviewer report is the primary source of substantive challenge in the revision. Peer feedback from the Referat is a secondary input – useful, but not load-bearing. If the assigned reader was weak or unfocused, that does not create a disadvantage, because the reviewer report supplies a well-calibrated challenge regardless.

Directional rotation for reader assignments

With n participants, the assignment is a single ring: person 1 reads person 2’s draft, person 2 reads person 3’s draft, and so on, with person n reading person 1’s draft. No one reads the draft of the person who is reading theirs. Pair-edit partners and Referat readers are deliberately different people.

The notes-handling rule

During each Referat, the writer takes notes on instructor-provided paper. At the end of each Referat session, the instructor collects the notes and holds them until week 14, when they are returned at the start of the supervised revision session. The rule prevents post-session expansion of the notes with an LLM – in the same logic as an exam booklet staying in the exam room.

Assessment

Grading happens against an assessment scheme with operational dimensions. Taste is the didactic orientation of the course; the assessment scheme is the graded instrument.

Essay (first-pass draft and revised draft)

Quality of the submitted text across its revisions:

Epistemic quality – correctness of technical claims, scope control, citation fidelity, handling of uncertainty and counterarguments.

Rhetorical quality – explanatory structure, clarity and concision, cohesion and coherence, register appropriate for an advanced undergraduate audience.

Document craft – figure quality and use, typographic discipline, visual hierarchy, overall polish and care.

Referat

Diagnostic quality – did the reader choose the right mode for the draft?

Substance of reading – does the reading engage with what the draft actually says?

Oral performance – pacing, structure, handling of discussion, handling of the on-the-spot editing question.

Revision and response letter

Quality of engagement with critique:

Effectiveness of revision – does the revised essay actually address the reviewer’s issues?

Quality of reasoning – is the participant engaging substantively with criticism, or performing engagement?

Preservation or improvement of the original essay’s strengths.

Weighting

30% essay (text quality), 30% Referat (quality of critical reading), 40% revision and response letter (engagement with critique).

Bonus points: the bad-paragraph bank

Throughout the semester, participants may submit examples of bad security or privacy writing found in the wild, each with one paragraph of annotation explaining what makes it bad. A good contribution earns a small bonus (up to 2 grade steps, e.g., from 2.7 to 2.0; capped at two contributions per person) and becomes part of the seminar’s teaching material in future years.

What makes superficial work harder

Several elements of the format raise the cost of shallow engagement:

• A participant who LLM-generates a draft without understanding it will struggle in the supervised revision session, because the reviewer report requires substantive engagement with claims in the text.
• A reader who rehearses a Referat with an LLM struggles with the on-the-spot editing question and with unscripted discussion.
• The directional rotation prevents pair-based collusion between readers and writers.
• The notes-handling rule prevents post-Referat LLM expansion of captured feedback.

Format at a glance

Topics are announced in week 1 and assigned in week 2. The first session is for experiencing the format in action.