146: Generative AI – Deeper Dive – 7-Part Livestream 2/7

❗️Is synthetic data trustworthy enough to train AI for patient care? It just might be—and that's what both excites and terrifies me. ❗️

Hey trailblazers! In this episode of the Digital Pathology Podcast, I take you through the second part of our AI in Pathology series—this time, we’re focusing on generative AI and how it’s revolutionizing diagnostics, education, and workflow in our field.

From synthetic H&E slides that could pass for real to multimodal agents that can read your histology images and chat with you about them—yes, really—this is where digital pathology meets the “bleeding edge” of AI development.

We’ll also look at real use cases, a synthetic biobank you can trust, and the biases, hallucinations, and ethical minefields that come along for the ride.

🔍 Highlights with Timestamps

• [00:00] Welcome + Paper Credits
  Kickoff and acknowledgments to the amazing authors behind the paper that inspired this series.
• [00:02] What is Generative AI?
  From rule-based systems to multimodal models and why everyone's suddenly Gen-AI.
• [00:04] AI Tools You Might Already Be Using
  ChatGPT, Claude, Gemini, Llama—and how they’re quietly shaping your workflow.
• [00:07] The Training Process Demystified
  Why it's not magic. From unsupervised learning to reinforcement with human feedback.
• [00:08] Retrieval Augmented Generation (RAG)
  How I use AI to update my book—and how you can use it to mine literature efficiently.
• [00:10] The AI Hierarchy Explained
  Where generative models fit into the bigger AI picture, including foundation models.
• [00:12] Synthetic Data in Practice
  Why training on rare diseases is finally realistic—and why that's a big deal.
• [00:15] Clinical Efficiency with Audio/Video AI
  How dictation and NLP are crushing administrative burden in healthcare.
• [00:18] Multimodal AI Use Case: PathChat
  Yes, I took a pic of my histology earrings—and yes, the AI knew what it was. 
• [00:25] Four Types of Generative AI in Medicine
  From simple text generators to complex multi-agent frameworks—examples and use cases.
• [00:37] Hallucinations, Biases, and Top-K Explained
  Understanding why LLMs “lie” and how we can manage it (temperature, top-P, and beyond).
• [00:45] Why Bias is Statistical, Not Personal
  Important talk on systemic bias in LLMs and geographic data interpretation.
• [00:47] Tech Talk: Building a RAG Pipeline
  What it really takes to use RAG effectively (it’s not just plug-and-play).
• [00:48] Risks, Regulations, and Tomorrow
  Benefits are massive, but privacy, cost, and consent still matter.
• [00:52] Course Announcement & Free Gift
  Get 50% off my structured AI in Pathology course using code VAL50.

📚 Resource from this episode

• 🔗 Main Reference Paper: ScienceDirect - Generative AI in Pathology