← Back to Contents
Note: The first drafts of these discussion questions were generated using Claude (Anthropic's AI assistant) and then reviewed and edited for the in-class use of this course.
Week 5 • Discussion

💬 Week 5 — Discussion Questions

To think about before class

These are example discussion points for you to think about before class. You are not expected to engage with all of them — pick the ones that speak most directly to your own research, and bring two or three rough answers to the in-class session. The full description of how to use these pages, including what the question tags mean, is on the Week 1 Discussion page.

Sub-lessons

The AI Literature Review Landscape

  1. Calibrate The lesson sorts AI literature tools into three categories. Take a tool you have used (Elicit, Semantic Scholar, NotebookLM, Claude, etc.) and place it in its category. What does the tool advertise as its strength, and where does its actual behaviour pull against the advertised strength?
  2. Apply Sketch what a combined workflow would look like for your own current literature-review needs. Which tool would do which job, and where do you hand work back to yourself for verification?
  3. Critical “AI changes literature review” is the lesson's framing. What hasn't changed — what part of literature-review work is, and probably will remain, irreducibly human?
  4. Connect Week 4 set out a personal framework for ethical AI use in research. Apply that framework to the literature-review workflow you would adopt off the back of this lesson. Which tool-use decisions does the framework comfortably endorse, which does it warn you against, and which does it leave ambiguous?

The Hallucinated Citation Crisis

  1. Calibrate Pick a recent paper or report claiming a hallucination rate for academic citations in AI tools. Read it against the verification habit the lesson asks for. Is the rate reported in a way you can act on, or is the number doing rhetorical rather than empirical work?
  2. Apply Pick the last AI-assisted reference list you produced (yours or a colleague's). Re-verify three random entries. Be precise about your verification protocol — what did you check, and what did you skip?
  3. Critical The NeurIPS 2025 scandal is presented as evidence that even careful researchers get caught. Is the right response (a) more verification, (b) different tools, (c) different incentive structures around what counts as a contribution? Pick one and defend it.
  4. Connect Week 2's training-data lesson made clear how much of the modern web is now plausibly the input to LLM training. This lesson makes clear how much of the modern literature is now plausibly the output. Read those two facts together: what does the “pollution loop” look like specifically for the literature in your own field, and what would “cleaning up” even look like?

Free Tools Deep Dive

  1. Calibrate Try the same well-defined literature-review task across two of the free tools covered. Where the results diverge, what does the divergence tell you about how each tool indexes, ranks, or interprets the literature?
  2. Apply Which free tool is most directly useful for a postgraduate researcher in your field? Be concrete about the tasks where it earns its place and the tasks where it doesn't.
  3. Critical Free tools come with structural constraints (rate limits, training-data cut-offs, hidden ranking biases). Which of those constraints would most affect your research, and how would you work around it without changing tool?
  4. Connect Week 3 made the energy and supply-chain costs of every AI query visible; Week 4 gave you a framework for the ethics of tool choice. Does your choice of free-tier literature-review tools follow consistently from those two weeks, or have you made the choice on cost grounds and then back-rationalised it?

Paid Tools and When They Are Worth It

  1. Calibrate Pick one paid tool (Elicit, Consensus, Scite.ai, SciSpace, etc.) and ask: what specific marginal capability would a 12-month subscription buy you over the free-tier alternatives? Put a number on it.
  2. Apply Sketch the case to a notional supervisor or department for one paid subscription. What does the case rest on — time saved, accuracy gained, or something else? Where does the case look strongest, and where would it be picked apart?
  3. Critical Paid tools advertise transparency about citation quality, scite-style support/contradict labels, etc. How much of that transparency is genuinely earned, and how much is interface polish over an underlying retrieval pipeline you can't inspect?
  4. Connect Week 3 made the per-query environmental costs of AI visible. Paid tools typically run on larger models per query than free tools. If you can do good research on free tools, are the paid versions a personal-productivity move, a quality move, or a footprint move — and which of those reasons would actually survive a careful audit?

Building Your Research Workflow with Claude

  1. Calibrate The lesson defines three levels (better prompting → Projects → Claude Code skills). Place your current Claude practice on that ladder. What would moving up one rung change about the quality, or just the speed, of your work?
  2. Apply Draft a CLAUDE.md (or its equivalent) for your own research context — the standing instructions you would want Claude to read before any session. What do you put in, and what do you deliberately leave out?
  3. Critical The lesson emphasises personalisation. Where might tightly-personalised setups (custom prompts, projects, skills) be quietly narrowing what you notice, by making the assistant agree with what you already believe?
  4. Connect Week 4 asked you to draw your own line between “AI as instrument” and “AI as collaborator.” Apply that line to the three levels in this lesson (better prompting → Projects → Claude Code skills). At which of the three levels do you cross from one side of your line to the other — and what does that say about how cleanly your line is drawn?

Hands-On Activities and Assessment

  1. Calibrate Compare your Comparative Search Challenge results. Where two tools surface the same papers in a different order, what does that tell you about how to read “most relevant” in any single tool's ranking?
  2. Apply Use the Citation Verification Exercise as the basis of a personal verification protocol you will follow on your next real literature review. What is the minimum number of checks that you actually intend to honour?
  3. Critical Hands-on assessment of literature-review tools tends to over-reward whichever tool you spend the most time tuning. How would you control for that bias in your own evaluation?
  4. Connect The verification protocol you keep up with here is the practical heir of the integrity-and-transparency norms from Week 4. Pick a step in your protocol you genuinely intend to honour and a step you suspect you will quietly drop under deadline pressure. What is the difference between the two, and does that difference square with the framework you committed to in Week 4?