32dots HEIDELBERG AI
Session 6 medium

Capstone: deep-read and reason through a real paper

LESSONLesson 6 · ~25 min

🎯Goal. Put it all together: upload a real paper and take Claude through it end-to-end — claims, methods, statistics, weaknesses — asking it to reason carefully and check itself before it concludes.

▶ Try this prompt

Here's the PDF of a paper in my field. Work through it carefully and reason step by step before answering: <questions> 1. State the central claim in one sentence, and quote the sentence where the authors make it. 2. Walk through their methods — does the design actually support that claim? Where are the gaps? 3. Examine the statistics: tests used, sample sizes, anything that should make me skeptical. 4. List the three weaknesses a tough reviewer would raise first. </questions> Before you give your final answer, re-check your reasoning against the actual text and correct anything you got wrong.

The trick that lifts quality on hard analysis: ask Claude to reason step by step AND to verify its own answer against the source before finishing. Combined with file upload (Lesson 2) and quoting, this turns a summary into a genuine critical read.

Steps
  1. 1Upload the paper and frame the whole task. Attach the PDF and give Claude the numbered set of questions above so it works through the paper as a structured critical read, not a quick summary.
  2. 2Ask it to reason, then self-check. Phrases like "reason step by step" and "re-check your reasoning against the actual text before you conclude" push Claude into deeper, more careful analysis and catch its own errors — especially valuable for stats and methods.
  3. 3Pressure-test and verify. Reply with "play the toughest reviewer — what's the single strongest objection?", then open the paper and confirm the quotes Claude used are real and in context before you trust any conclusion.

You'll see. A structured, quote-backed critical reading of a real paper — central claim, method gaps, statistical red flags, and the objections a hard reviewer would raise — with Claude having checked its own reasoning along the way.

💳Cost. Everything here runs on the free plan; a long, multi-question deep-read will use up the free daily message limit quickly, where Pro/Max simply let you keep going. Careful step-by-step reasoning works best on Claude's strongest models, available across the plans.

💡Takeaway. For your hardest reading, combine the whole course: upload the paper, ask Claude to reason step by step and verify itself against the text, then check the quotes yourself — that's a critical read you can actually defend.

How was this lesson?