Big Technology Podcast AI's Research Frontier: Memory, World Models, & Planning — With Joelle Pineau
239 snips
Feb 4, 2026 
Joelle Pineau, Chief AI Officer at Cohere and long‑time AI researcher, discusses frontier research in memory, world models, and efficient hierarchical planning. She contrasts memory versus continual learning. She explains retrieval challenges, distinctions between digital and physical world models, and why many specialized agents — not one monolithic system — are likely to emerge.
AI Snips
Chapters
Transcript
Episode notes
Three Core AI Research Frontiers
- Memory, world models, and efficient reasoning are the three top research frontiers Joelle Pineau highlights.
- Advances must target selective retrieval, predicting action effects, and hierarchical planning across time scales.
Memory Versus Continual Learning
- Memory and continual learning are related but distinct problems according to Joelle Pineau.
- Memory focuses on selective retrieval and efficiency while continual learning emphasizes non-stationarity and changing data.
Don't Let Models Learn Unchecked
- Avoid enabling unchecked online learning for deployed models without strong testing safeguards.
- Test models before release and prefer staged updates to prevent rapid undesirable behavior shifts.