Delivered on Day 3 of the CAITG AI Winter School (Session 2, “AI for Literature Review”), this lecture (“Breaking Down the Literature Review: what to offload to an LLM, what to keep for yourself”) treats the literature review as an ongoing process rather than a one-off task, and asks where AI agents genuinely help versus where they fall short on citation accuracy, provenance and reproducibility.

It works through:

  • Why break the review into tasks at all, and a caveat about doing so.
  • The six tasks — problem formulation, literature search, screening for inclusion, quality assessment, data extraction, and data analysis and interpretation — with a judgement on each about what to delegate to a model.
  • Note-taking as distillation — the Zettelkasten move of keeping a map of the collection rather than just the cards, what is worth retaining beyond the note itself, and single-shot versus interactive extraction.
  • Connecting ideas across papers — and how to make paper-to-paper connections actually trustworthy.