Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
AI in Real Life: Using LLMs to Turbocharge Microsoft Learn
Summary
Enthusiasm for AI tools, especially large language models like ChatGPT, is everywhere, but what does it actually look like to deliver large-scale user-facing experiences using these tools in a production environment? Clearly they're powerful, but what do they need to make them work reliably and at scale? In this session, Sarah provides a perspective on some of the information architecture and user experience infrastructure organizations need to effectively leverage AI. She also shares three AI experiences currently live on Microsoft Learn: An interactive assistant that helps users post high-quality questions to a community forum A tool that dynamically creates learning plans based on goals the user shares A training assistant that clarifies, defines, and guides learners while they study Through lessons learned from shipping these experiences over the last two years, UXers, IAs, and PMs will come away with a better sense of what they might need to make these hyped-up technologies work in real life.
Key Insights
-
•
Most AI applications no longer require building foundation models from scratch; the focus is now on application development and integration.
-
•
Single, all-purpose chatbots (everything chatbots) are insufficient because they handle high ambiguity and diverse, often complex tasks poorly.
-
•
Sarah introduces the ambiguity footprint as a framework to measure AI application complexity and risks across several axes such as task complexity, context, interface, prompt openness, and sensitivity.
-
•
AI features that support simple, complimentary user tasks, rather than critical or complex ones, are easier and safer to build and scale.
-
•
Visible AI interfaces, like chatbots, set clearer user expectations but introduce more ambiguity and management overhead compared to invisible AI (e.g., keyboard optimizations).
-
•
Prompt engineering plays a crucial role in defining the boundaries of AI output, from very open-ended to highly restricted scopes.
-
•
Retrieval Augmented Generation (RAG) helps manage up-to-date context by dynamically querying relevant data chunks rather than using static corpus.
-
•
Evaluating AI outputs rigorously is essential but often underprioritized; without clear quality metrics, teams end up relying on subjective or anecdotal assessments.
-
•
Data ethics and distributed AI implementations can create blind spots, limiting feedback loops necessary for continuous AI model improvement.
-
•
Incrementally building AI applications with smaller ambiguity footprints helps organizations develop expertise and controls before tackling more complex, open-ended AI products.
Notable Quotes
"You’re not doing IA, but you’re always doing it."
"An everything chat bot is almost certainly not how you’re going to build it; realistically you’re building three apps in a trench coat."
"AI is ambiguous at best because we’re fully in the realm of probabilistic rather than deterministic programming."
"The more complex the task, the less likely it is to be successful with current AI."
"A task where AI adds a little something is honestly easier to get right than one where it’s absolutely critical."
"Visible AI interfaces introduce another place where you can add ambiguity."
"Retrieval Augmented Generation lets you supply specific relevant information to the model dynamically rather than everything at once."
"Evaluation might be the most important part of your entire development effort and is often the hardest to do well."
"You can’t just eyeball results and call it good; AI applications are expensive and complex and require systematic evaluation."
"Never build or buy an everything chat bot again; start with less ambiguous, targeted AI experiences."
Or choose a question:
More Videos
"Our technical leads started attending other teams’ meetings, helping break down integration barriers."
Carl TurnerYou Can Do This: Understand and Solve Organizational Problems to Jumpstart a Dead Project
March 28, 2023
"Don’t send your recruiters to spam LinkedIn. That damages your brand and turns off potential candidates."
Karen PascoeDeveloping Experience Teams and Talent in the Enterprise
June 8, 2016
"Support software used to be really shitty and expensive, and Zendesk set out to fix that by making it beautifully simple."
Jilanna WilsonDistributed Design Operations Management
October 23, 2019
"Reliable access to veterans and other users is fundamental to running quality, inclusive research across many teams and products."
Shawna Hein Kevin HoffmanCreate a Cohesive Civic Design Practice Across Agency, Vendors, and Contracts
November 17, 2022
"I still kind of pause in my tracks when I see our corporate vision: a world without fear of cancer."
Barb SpantonDoing Work That Matters: A Look Beyond The Idealistic Notion of 'Doing Meaningful Work'
June 10, 2022
"We had no recipe, no formula, just vibes."
Briana ThomasThe Quiet Force: Uncovering Hidden Leadership in High-Impact Design Teams
September 24, 2024
"Service designers connect the dots, which represent elements, teams, processes, policies, and customer needs that all must align."
Gina MendoliaTherapists, Coaches, and Grandmas: Techniques for Service Design in Complex Systems
December 3, 2024
"Now everyone’s an edge case, so research needs to focus on those edge cases, not just the average user."
Lija Hogan Milan Mijatovic Sam Proulx Louis RosenfeldThree Years Out: Perspectives on the Near-Term Future of User Research
March 15, 2024
"DPM loneliness and siloing happen because of wearing too many hats and inheriting mismatched responsibilities."
John Calhoun Rachel PosmanTwo Sides of the DesignOps Coin: Teams Ops and Product Ops
January 8, 2024
