The three of us talk monthly in the Libraries Lead podcast (available at librarieslead.libraryjournal.com), and share content from that segment of the podcast in digital and print form through Library Journal. In this month’s column we talk about:

• Trapped in a ChatGPT Spiral
• Cloudflare CEO Matthew Prince on how AI could "ruin the internet"
• Disaster preparedness and using AI tools to recreate websites
• Claude AI
• The recent Anthropic copyright settlement

David Lankes
University of Texas at Austin

I have two items to contribute this month. The first, an immediate concern, stems from the NY TImes podcast, The Daily, episode “Trapped in a Chat GPT Spiral” (September 16). This episode details how people, especially youth, are engaging with AI as a friend, therapist, or romantic partner. Continuing these intense, isolated conversations can lead to dangerous delusions—making users feel brilliant or uniquely enlightened—and in tragic cases, has been linked to suicide. Beyond simple hallucinations, the episode warns about AI’s ability to create delusive “bubbles" that isolate and mislead users.

The second, a future-thinking concern, comes from a Gizmodo article on Cloudflare CEO Matthew Prince’s forecast for how AI could “ruin the internet.” Prince argues we’ve shifted from search engines to “answer engines” (like the top box on Google or a direct ChatGPT response) that don’t credit original sources. This is causing a dramatic drop in website traffic and ad revenue, reducing the incentive for humans to create quality content.

Prince outlines three scenarios:

1. The “Internet of the Dead”: The web becomes a wasteland of bots creating content for other bots, with human agency vanishing.
2. AI Patronage: Massive AI companies fund and hire content creators—journalists, artists—to produce the necessary human-generated content, making them the ultimate gatekeepers.
3. Licensing and Payback (Prince’s preference): A system is created to force AI models to license and pay for the data they use for training. Since Cloudflare sits as a middleware provider, Prince suggests they and others can technically block un-licensed AI bots from accessing content.

The core challenge is finding a viable path to sustain human-created content and ensure its value in a world dominated by sophisticated AI answers.

Beth Patin
Syracuse University

My IMLS grant got reinstated! I’m now working with a great group of student workers to build a website for my disaster preparedness starter kit for libraries. Working with information management students has been eye-opening. They’re all in on AI tools, while my library and PhD students are more like me: a bit more critical and skeptical. We’re using Figma to build the site, and it has this AI chat feature. During a meeting, I asked if I could play around with it. My student said yes, so I tried it out. I asked the tool to recreate the website in Syracuse colors without telling it what those colors actually were. The AI went out and found our exact color reference numbers and branding guidelines. It worked really well.

This experience made me realize I should challenge myself to play with these tools more and think about how I might actually use them. Syracuse just announced that all faculty and students are getting access to Claude AI, which was developed by Anthropic. What’s supposedly different about Claude is its “constitutional AI” approach. It uses predefined ethical rules instead of just human feedback to guide decision-making, which is meant to make outputs safer and more reliable.

Claude can do things like generate creative content, summarize documents, brainstorm ideas, and analyze PDFs and Word files—something ChatGPT can’t quite do the same way. I‘m going to spend the next month exploring Claude, and I'll report back next time with what I find.

Mike Eisenberg
University of Washington

I looked into the Anthropic settlement, which could become the largest copyright settlement in U.S. history. Anthropic, one of the top AI companies along with Open AI, has agreed to a settlement of $1.5 billion to a group of authors and publishers after a judge ruled that it had illegally downloaded and stored millions of copyrighted books. Anthropic is supposed to pay $3,000 per work to over 500,000 authors. It sounds good at first read or listen, but it's actually not as momentous as I would hope for.

The judge actually ruled that Anthropic was within its rights to acquire and use copyrighted books to train their AI bot as long as they had bought those books and other works legally. That’s fair use.

But Anthropic was at fault because they had also used pirate-type services such as Library Genesis and Pirate Library Mirror to illegally to download content they knew contained pirated works. After the judge ruled that Anthropic was going to be held accountable for the pirated works, Anthropic worked out a settlement with the authors and publishers, preliminarily approved on September 25.

To summarize, the Anthropic settlement should not be misread as a referendum against AI training. Rather, respect copyright, aquire works legally, and you can use them to train your AIs.