BETH PATIN:

Testing Claude: A Skeptic's Initial Assessment

As perhaps the most AI-skeptical member of our podcast trio, I've been hesitant to adopt generative AI tools in my work. However, when Syracuse University secured access to Claude, Anthropic's constitutionally-aligned AI designed to make ethical decisions when sharing information, I challenged myself to give it a fair trial.

For my first test, I leveraged Claude's document analysis capabilities with materials I knew well: the LIS Forward Project papers from the University of Washington that I co-authored with Dave and many others from the LIS world. I uploaded the PDFs and asked Claude to generate talking points for my dean about strengthening our MLIS program and aligning our curriculum with what library directors seek in new graduates. The results were impressive. Claude produced thoughtful, actionable recommendations that accurately reflected the field research we'd gathered. Because I was deeply familiar with the source material, I could verify the quality of its analysis, and it held up under scrutiny.

My second experiment was less successful. When I uploaded my calendar file seeking optimal times for recurring board meetings, Claude earnestly suggested Saturdays and Sundays, technically my only open slots, but hardly practical for work commitments. Even with additional parameters, it struggled to understand the nuance of work-life boundaries.

Current score: 7/10. Claude excels at document analysis and insight synthesis but falters at contextual decision-making. I'll continue experimenting to see where else it proves useful, or falls short.

References:

LIS Forward. (2023). Ensuring a Vibrant Future for LIS in iSchools: The Friday Harbor Papers, Volume 1. University of Washington Information School.  https://doi.org/10.6069/TG4M-8B21 
LIS Forward. (2025). Responses to the LIS Forward Position Paper: Ensuring a Vibrant Future for LIS in iSchools, The Friday Harbor Papers, Volume 2. University of Washington Information School.  https://doi.org/10.6069/F6CQ-H317

Roberts, L. H., & Dean, J. W., Jr. (2025, October 9). University leadership announces development of new school at Carolina. The University of North Carolina at Chapel Hill. https://www.unc.edu/posts/2025/10/09/development-of-a-new-school-at-carolina

Goelz, A. (2025, September 22). SU to offer Anthropic’s ‘Claude for Education’ campus-wide. The Daily Orange. https://dailyorange.com/2025/09/su-offer-anthropic-claude-for-education-campus/

DAVE LANKES:

OpenAI rolled out a text-to-video model called SORA in 2024. They’ve just released SORA 2, and honestly, it’s a bit depressing. They didn’t just release a new text-to-video model—they released it as a social media app. Now you can not only generate videos, you can instantly blast them across the internet.

When you sign up, it tells you to look right, look left, and then say three numbers. From those 10 seconds it builds something called a cameo. That cameo can be used by anyone (with your permission) to drop “you” into the AI generated videos. The videos it produces are remarkably good, but creepy and wrong on so many levels. And now people can churn out videos with you in them.

OpenAI pitches Sora 2 as a ‘creativity engine,’ full of fun and possibility. But honestly, if you watch the promo video, it’s also a spectacular, high-gloss demonstration of how easy it now is to produce misinformation and deep fakes.

It’s incredible tech. But OpenAI is very clearly building it to make it easy to create—let’s be blunt—deceptive content, all under the banner of creativity.

OpenAI. (2025, September 30). Sora 2 is here. OpenAI. https://openai.com/index/sora-2

MIKE EISENBERG:

How well AI is doing? Answer – not so well! An MIT NANDA (Networked Agents and Decentralized AI) Initiative study found that in businesses, 95% of AI pilot implementation projects fail!  That’s right, only 5% are successful. The 2025 MIT NANDA Report (2025) analyzed over 300 public AI deployment, 52 organizational interviews and surveys from 153 leaders.

For generative AI, 95% of the projects fail to deliver measurable business value. It’s not because of the models; the algorithms are great. It's because of poor integration, vague goals, and organizational resistance to “friction.” In most companies, AI is often dropped into the workflow without any purpose, without integration, without buy-in from the people who actually use it.

AI teams tend to build in isolation. They don't involve frontline users. They don't gather feedback and they don't adapt the tools to real world needs. The result, low adoption and minimal impact. And, perhaps most importantly, companies try to avoid friction. They're afraid to disrupt existing processes. Again, the problem is not the algorithm. The problem is the human and organizational side.

So, Gen AI is often an add-on and creates problems. What are some implications for libraries and librarians? I thought of two right off the bat. One, to those involved with AI implementations, focus on the organizational aspects and not just what the AI does. Second, librarians can help businesses and organizations in their communities to better use and implement AI with the same message: emphasize the organizational factors and context, the people and processes affected, and integration, not add-on.

Challapally, A. (2025). The GenAI divide: State of AI in business 2025. MIT NANDA Initiative. https://nanda.mit.edu/reports/genai-divide-2025