The Data Exchange with Ben Lorica Reading the Tea Leaves: What the World's Top AI Researchers Are Really Working On
14 snips
Apr 30, 2026 Nick Vasiloglou, VP of Research at Relational AI and conference data analyst, digs into NeurIPS trends. He highlights data and model markets, the rise of small but powerful LLMs, large context windows and ring attention, AI accelerating scientific discovery, and tighter integration of models with databases and enterprise data. Short, focused takes on what to watch next in AI research.
AI Snips
Chapters
Transcript
Episode notes
Attribution Unlocks Practical Data Markets
- Data markets are resurging because new attribution methods can trace model outputs back to specific training examples.
- Recent papers sped up influence computations so attribution can be practical in real time for text and images.
Real Time Attribution Is Now Achievable
- Attribution tech matured from slow prototypes to systems fast enough for real-time use, enabling copyright and payout mechanisms.
- Nick cited examples tracing an answer to a sci‑fi novel in another language as proof of concept.
Model Markets Through Composition And Routing
- Model composition and routing (mixture of experts) is emerging as a practical way to build large systems from smaller specialist models.
- Cohere and others trained distributedly and combine weights or route to specialists instead of monolithic trillion-parameter runs.