What happens when you mix live faculty with generative AI? (www.sampaddock.com)

Berkeley-hosted Generative AI 1-Week Intensive debuted a new hybrid pedagogy that pairs live faculty (a “Dream Team” of academics and industry experts) with an embedded AI tutor, Paski. Run to 130+ global participants, the pilot produced strong outcomes: NPS +78 and learners spending 50% (9 of 18 hours) of course time with Paski. Students used voice for 89% of interactions and asked to “keep Paski forever,” underscoring high engagement and perceived utility. The program produced work-ready AI strategy documents by applying DAIN’s 9-part framework, blurring education and consulting for professionals. Technically, Paski is an orchestrated agent built into the course flow rather than a peripheral add‑on: it uses OpenAI’s Realtime API, contextual prompts grounded in course materials, student-profile variables, dynamic tool calls (document/slide/whiteboard loading, transcript retrieval, agent chaining), and ongoing dialogue memory. The campus integrates live video + instant transcripts, faculty “digital twins,” and personalized audio debriefs (NotebookLM audio) that tailor lecture summaries to each learner’s role. Key lessons for the AI/ML community: tutor adoption requires deep integration and UX choices (e.g., student-driven voice control), targeted personalization (~5–10% of core curriculum) boosts relevance, and agentic tool-chaining enables AI to act as both tutor and consultant—pointing to scalable models for high-engagement, work-applied learning while highlighting areas for further research (longitudinal tutor persistence, privacy and pedagogy trade-offs).