Beyond The Prompt - How to use AI in your company AI’s Next Frontier: World Models Explained by Christian Keller
12 snips
Nov 27, 2025 Christian Keller, Product Lead at Meta's Superintelligence Lab and seasoned AI expert, dives into world models and their pivotal role in generative AI. He explains how AI benefits from video input, enhancing understanding of cause and effect that text misses. Christian shares insights on evolving workflows, encouraging listeners to ask, 'Could AI help with this?' He illustrates this with personal anecdotes, including how his wife creatively used generative tools for French exercises. The discussion also touches on tackling bias in models and reimagining organizational roles with AI.
AI Snips
Chapters
Transcript
Episode notes
Text Loses What Video Restores
- Text compresses reality and loses context, detail and temporality.
- Images and especially video restore context and show change over time for richer learning.
Validate With Generic Models First
- Start product experiments with the best available generic model and iterate with prompts and system messages.
- Once you validate fit, invest in fine-tuning, benchmarks and a data strategy to specialize performance.
LLMs Hallucinate; World Models Predict
- LLMs predict likely next tokens and therefore hallucinate by design.
- World models that predict state changes (via video) give more robust cause-and-effect understanding.