Summary
Keeping large content repositories organized is an ongoing challenge. There's always new stuff coming in, and taxonomies evolve over time. Resource-strapped teams seldom have opportunities to re-categorize older content. It's a task well-suited for generative AI. Large language models have powerful capabilities that can help teams keep content organized at scale. Using LLMs in this capacity can lead to better user experiences and free team members to focus on more valuable efforts. This presentation explores two approaches for using LLMs to organize content at scale: 1) re-categorizing content using existing categories and 2) developing new categories from existing content. Both will be shown as proofs of concept alongside feasible next steps.
Key Insights
-
•
Large content repositories often become disorganized over time as products and markets evolve, yet organizations tend to deprioritize reorganizing content.
-
•
Search alone is insufficient for navigating large content repositories because users may lack context about the content available.
-
•
Taxonomies and content categories help users understand and navigate both external and internal content repositories.
-
•
AI, particularly large language models like GPT-4, can effectively assist in retagging large amounts of existing content faster than manual processes.
-
•
Maintaining human oversight while using AI tools is critical to prevent errors such as hallucinated tags or irrelevant categorizations.
-
•
Creating new taxonomies for large repositories requires analyzing the entire corpus, for which techniques like embedding databases and retrieval augmented generation (RAG) are useful.
-
•
Graph RAG, which integrates knowledge graphs with LLMs, improves precision by incorporating semantic relationships beyond keyword similarity.
-
•
AI tools enable new workflows that increase not just speed but open novel possibilities in content organization and information architecture.
-
•
Privacy concerns in client work lead to experimentation with local AI models to avoid exposing sensitive internal content.
-
•
Learning AI tools hands-on is essential for content professionals to harness their full potential and adapt workflows accordingly.
Notable Quotes
"Keeping large content repositories organized is an ongoing challenge that often gets deprioritized."
"Search alone might not cut it, especially because people often don’t know what to look for or don’t have enough context."
"I thought, if terms aren’t understandable to users, GPT-4 probably won’t understand them either, so I cleaned up the taxonomy."
"The LLM introduced some tags of its own even though I asked it not to, so I made sure to review everything before changes went live."
"Using GPT-4 to retag 1,200 posts took about a third of the time manual tagging would have taken."
"LLMs are optimized to work with short snippets of text, so we have to prepare content carefully before feeding it to them for big picture analysis."
"Graph RAG improves on plain text RAG by using knowledge graphs, adding semantic relationships that boost precision."
"These tools have increased both my speed, efficiency, and efficacy as an information architect."
"Working with these tools requires developing different workflows and questioning how we’ve always done things."
"The only way to get a taste for what these AI tools can do is by actually getting stuff done with them."
Or choose a question:
More Videos
"A 101 conversation over coffee yields better insights than written reviews in 360 assessments."
Adam Cutler Karen Pascoe Ian Swinson Susan WorthmanDiscussion
June 8, 2016
"Sometimes the director is the most senior design person in the org and ends up playing the executive role without the title or support."
Peter MerholzThe Trials and Tribulations of Directors of UX (Videoconference)
July 13, 2023
"In the automotive industry, most safety standards only became mandatory after governments intervened."
Lisa WelchmanCleaning Up Our Mess: Digital Governance for Designers
June 14, 2018
"Investing in sustainability today will yield dividends for future generations."
Vincent BrathwaiteOpener: Past, Present, and Future—Closing the Racial Divide in Design Teams
October 22, 2020
"It matters what you build, but it matters more if you learn."
Brenna FallonLearning Over Outcomes
October 24, 2019
"Hip hop proves that we can re-animate spaces with highly generative communities that weren't built for us."
Tricia WangSpatial Collapse: Designing for Emergent Culture
January 8, 2024
"Diverging and converging around the business model canvas helped us test and prototype delivery methods for the value propositions."
Edgar Anzaldua MorenoUsing Research to Determine Unique Value Proposition
March 11, 2021
"Our brains are terrible at operating only on one type of information, whether object-oriented or context-oriented."
Designing Systems at Scale
November 7, 2018
"Randomization is magic — it evenly distributes confounds so the only difference affecting results is your change."
Erin WeigelGet Your Whole Team Testing to Design for Impact
July 24, 2024
