AI Skill Report Card

Converting Research Papers to Presentations

A-83·Sep 29, 2026·Source: Extension-page
14 / 15

Given a research paper PDF, produce a slide outline that maps academic sections to slide layouts without inventing content:

Slide 1: Title — paper title, authors, affiliation, venue/date
Slide 2: Motivation — problem statement + why it matters (from Intro)
Slide 3: Abstract Summary — 3-4 bullet condensation of the abstract
Slide 4-5: Methodology — approach, dataset, experimental setup (diagrams if present)
Slide 6-8: Results — key findings, one major result/figure per slide, exact numbers preserved
Slide 9: Discussion — interpretation, limitations
Slide 10: Conclusion — contributions + future work
Slide 11: References — key citations only (3-6 most load-bearing)
Recommendation▾
Add a bad-example contrast (e.g., a poorly compressed results slide vs. a good one) to strengthen the examples section
14 / 15

Progress:

  • Step 1: Parse the paper's structure (identify section boundaries: Abstract, Intro, Methods, Results, Discussion, Conclusion, References)
  • Step 2: Extract verbatim key facts — statistics, formulas, dataset names, exact figures/tables — flag these as "do not paraphrase"
  • Step 3: Condense narrative sections (Intro, Discussion) into 3-5 bullets max per slide
  • Step 4: Map each section to an appropriate slide layout (title, bullets, figure+caption, comparison table, big-number stat)
  • Step 5: Preserve figures/tables by referencing their captions; don't fabricate data if a figure can't be reproduced — describe it textually instead
  • Step 6: Draft speaker notes per slide with the fuller context that didn't fit on-slide
  • Step 7: Review for length — one paper section should not sprawl across too many slides; aim for 10-16 slides for a standard 10-15 min talk
  • Step 8: Output outline (or hand off to slide-generation tool) and flag any low-confidence extractions for human review
Recommendation▾
Include guidance on handling papers with unusual structures (e.g., no clear Methods section, preprints, or multi-study papers)
15 / 20

Example 1: Input: A 12-page single-column ML conference paper with Abstract, Related Work, Method, Experiments, Ablations, Conclusion. Output: 13-slide deck — Title, Motivation (from Intro gap), Related Work (1 slide, 3 bullets), Method (2 slides: architecture diagram + algorithm steps), Experiments setup (1 slide: datasets/baselines), Results (2 slides: main table reproduced with exact numbers, one chart described), Ablations (1 slide), Limitations (1 slide, pulled from Discussion/Conclusion), Conclusion + Future Work (1 slide), References (1 slide, top 5 citations).

Example 2: Input: A two-column journal article in biology with dense statistical results (p-values, effect sizes). Output: Results section becomes 3 slides instead of 1, each anchored on one hypothesis/finding, with exact p-values and effect sizes quoted rather than rounded or generalized. A note flags that one figure (a complex multi-panel plot) should be inserted as an image rather than recreated.

Recommendation▾
Provide a concrete speaker-notes example to illustrate Step 6 rather than just describing it abstractly
  • Treat numbers, quotes, and statistical results as sacrosanct — copy exactly, never round or reinterpret.
  • One key idea per slide; split dense sections (especially Results) rather than cramming.
  • Keep academic flow (Abstract → Methods → Results → Conclusion) — don't reorder for "narrative flair."
  • Use figure/table captions as-is when summarizing visual content you can't reproduce.
  • Distinguish paraphrased content (safe to compress) from quoted/quantitative content (must stay exact).
  • Include a references slide with only the most essential citations, not the full bibliography.
  • For two-column or non-standard PDF layouts, verify section detection didn't interleave columns incorrectly.
  • Don't summarize the Results section as vague prose ("the method performed well") — always retain concrete metrics.
  • Don't merge Methodology and Results into one slide; they serve different audience questions.
  • Don't fabricate or estimate data points for figures that can't be extracted — describe them qualitatively instead and flag for manual insertion.
  • Don't drop limitations/discussion content just because it's not "positive" — it's part of academic integrity.
  • Don't over-compress the abstract into a single line; it should still convey problem, method, and result.
  • Avoid slide walls of text — if a bullet exceeds ~20 words, split or trim it.
0
Grade A-AI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
14/15
Workflow
14/15
Examples
15/20
Completeness
17/20
Format
15/15
Conciseness
13/15