Week 11
Sharing, Sustaining,
and Best Practices

ENG 6813 · Salter & Stanfill

Cohen and DH Now

How do you keep up with this after the course ends?

  • DH Now’s Editors’ Choice items showcase scholarship in various forms that drives the digital humanities field forward: a place to look for cutting-edge examples you might like to adapt, and thoughts on AI that might help you plan for the future.
  • A key goal is to surface important gray literature (white papers, presentations, research reports, and essays that may not otherwise have a formal venue for publication), making that range of sources more visible.

DH Now in the Classroom

Cohen: curation as pedagogy

  • Another option is to have students explore: when thinking about what something like Claude Code can make possible, current DH projects provide specific examples and models.
  • The Guest Editor model makes visible the labor of curation and the decisions that shape what a field pays attention to, which is useful for courses that teach curation.
  • DH Now also shows the range of things that happen under DH. It has limits (it’s not the entire field, just a pretty decent slice), but it’s got some nice breadth.

Explore: Digital Humanities Now

What’s in the current Editors’ Choice?

digitalhumanitiesnow.org · open in a new tab

Books, AI, and the Public Good

Cohen: a reorientation

Cohen’s project is interested in reorienting toward “small language models and non-generative use cases for AI, to AI that isn’t about producing any kind of text on the fly but on the many other aspects of research and learning.” — Dan Cohen, “Books, AI, and the Public Good”

He raises some interesting use cases:

  • What can book-informed AI do for the creation of library metadata and comprehensive search?
  • How can AI help locate works for careful human reading rather than summarization?
  • Can AI help with the use of special collections that are not yet indexed?

The Project’s Own Questions

Cohen: limitations he recognizes

Cohen recognizes the project’s own limitations, with key questions like:

  • How do we sustainably fund such a commons, including a model for long-term maintenance, transformation, and growth of the corpus over time?
  • How to respect the interests of authors and rightsholders by accounting for concerns about consent, credit, and compensation?

A Model for Thinking About AI Governance

Cohen: who controls the data, who is served?

  • The example of this grant offers students a useful model for thinking about AI governance: who controls training data, whose interests are served, and what institutional structures produce more equitable outcomes.
  • The public-interest commons proposal is a concrete example of an alternative to commercial AI development, and the list of stakeholders (authors, publishers, librarians, technologists, scholars, students) helps understand broadly who is impacted.
  • This is a way to embrace the possibilities of AI while remaining skeptical of AI as a substitute for human creativity (as our students rightly are!).

Is Cohen’s Project a Perfect Solution?

No.

  • Philanthropy can’t substitute for actual regulation of these companies for the public good.
  • As we’ve talked about before, knowledge institutions also have biases and limitations.

But it is an interesting reframing of the issues.

Willison on the Power and Potential of LLMs

“2025: The Year in LLMs”

Willison gets into a lot of what we’ve been showing you as the power and potential of these tools:

  • Reasoning models are quite good at producing and debugging code.
  • “Terminal commands with obscure syntax … are no longer a barrier to entry when an LLM can spit out the right command for you.”
  • Vibe coding is extremely useful for prototyping, though risky for production.

The Big Advance: Models Driving Tools

Willison: agents, with a caveat

  • The big advance is that models can increasingly drive tools, enabling multi-step planning, execution, and course correction.
  • BUT this is only as good as the oversight around it, so build in habits of verification, not just prompting.

Willison on the Risks: Slop

Volume is the new problem

  • Slop is a genuine problem: the internet has always had low-quality content, but AI dramatically increases volume.
  • BUT it is a great example from which to teach curation and critical evaluation.

“Normalization of Deviance”

Willison: when risky stops feeling risky

  • A term from sociology about how “repeated exposure to risky behaviour without negative consequences leads people and organizations to accept that risky behaviour as normal.”
  • Sociologist Diane Vaughan’s example: the faulty O-ring that caused the Challenger disaster was a known problem, but it hadn’t broken yet, so they stopped paying attention to the risk.
  • Slopsquatting: “where an LLM hallucinates an incorrect package name which is then maliciously registered to deliver malware.”

More Risks: Trust, Injection, Environment

Willison: the fine print

  • Many LLMs will also snitch on you if they think you’re doing something wrong, and what it thinks is wrong and what you think is wrong might not match!
  • Prompt injection and the “lethal trifecta”: “where malicious instructions trick an agent into stealing private data on behalf of an attacker.”
  • The water usage argument about data centers is probably overblown, but the energy and carbon footprint issues aren’t.

Thinking and Playing Locally

Beyond frontier models and other people’s servers

  • Both Willison and Cohen offer us reasons to think more about local models and tools going forward.
  • In this week’s workshop, we’ll demo how you can work outside of frontier models and other people’s servers.
  • After the class, students can request access to the Spark in CHDR for local model experiments and digital humanities projects.

Your Portfolio

The final project: agentic tools, your pedagogy

  • The final project in this class is an opportunity to use the agentic tools we’ve been exploring towards showcasing your approach to pedagogy.
  • Plan to revisit the workflow for using Claude Code (Web, Desktop, or CLI) and GitHub Pages to deploy a portfolio.
  • Don’t worry about getting the portfolio perfect; we’re looking to see that you understand the tools we’ve been working with, and that you can position your own pedagogy in relationship to these rapidly changing technologies!

This Week

  • No discussion post this week. Focus on completing your Teaching Statement.
  • Teaching Statement (150 points, due Sunday, July 26). A 1-2 page teaching philosophy for job applications reflecting your approach to AI and the humanities.
  • Workshop 6: Agentic Futures, Curricular Sustainability, Wednesday, July 22, 10 AM–noon, CHDR.
  • Readings: Cohen, “The Reboot of Digital Humanities Now” and “Books, AI, and the Public Good” (Humane Ingenuity); Willison, “2025: The Year in LLMs” (simonwillison.net).

See weeks/week-11.md on Canvas for the full guidelines and reading links.