Critical Thinking with AI • Hist 300

Trusted Sources: A Guided Source Hunt

Think of this like tracking an animal, not ticking a checklist. Today we ask a different question: not is it good or bad? but what is this source fit for?

Provocation

Not all sources are equal — but all are useful for something.

Reddit, Wikipedia, an academic blog, and a peer-reviewed book or article provide not better or worse information, but different kinds of insights. The goal is not good vs. bad. The goal is fit for purpose.

What Counts as a Historical Source

Evidence vs. Argument About Evidence

Primary — Evidence

Created in the moment: letters, speeches, census data, photos, oral histories.

Secondary — Argument

Interpretations of the past: books, scholarly articles, reviews.

We're focused on secondary sources as a synthesis of historical interpretation — but remember, they are still sources that need to be evaluated!

Use diverse search strategies

Cast wide nets. Aim small.

Using Google means finding lots of driftwood quickly. USEFUL! Don't neglect slower and more curated tree hunting. USEFUL!

01

Library databases

JSTOR, Project MUSE — curated scholarship with real vetting.

02

Archival collections

Library of Congress, National Archives — primary evidence at scale.

How to Evaluate: Four facets

View every source under four kinds of lenses:

01 · Authorship

Who made this?

What is their expertise, stake, or position?

02 · Context

When and why?

What moment produced it, and what was it responding to?

03 · Evidence

Based on what?

What is this built from — data, testimony, other sources?

04 · Specificity

On target?

Potentially too distracting or a valuable new perspective?

History isn't just facts — it's interpretation shaped by context. Be broad but selective with your contexts

The Trap of Credibility

Stop asking: is it credible?

This is relevant only if you believe in absolute truth. For AI purposes, we want to ask instead: what kind of truth is this source capable of telling?

  • A primary source can be biased, incomplete, or deliberately misleading.
  • A peer-reviewed article can still be wrong, dated, or contested.
  • Replace the question with: what kind of truth is this source capable of telling?
  • How can this guide the LLM's understanding and broaden its perspectives?

Corroboration

Interpretation hinges on triangulation.

No single source or perspective stands alone. Historians build arguments by stitching camera angles into a scene — and noticing where the angles can tell different stories.

  • Look for agreement — what do multiple sources confirm?
  • Look for disagreement — where do they conflict, and why?
  • Look for silence — what is missing, unrecorded, or unspeakable?

Closing Takeaway

Finding sources is assembling perspectives, not locating truth.

The real skill is the chain: question → source → interpretation. That's the process you need to verify with AI. It's good at faking it!

Practice →

Now let's try this in NotebookLM.

We'll load a small set of sources, ask a real historical question, and watch how question, source, and interpretation push on each other in practice.

  • Start with a basic question.
  • Think about these slides.
  • Triangulate — agreement, disagreement, silence.
  • Formulate — what story emerges? what does it matter?