John Platt (Google): AI for Climate and Science
Oct 9, 2026
The Information Bottleneck
Kaggle Grandmasters, Agent Skills, and Why Everyone Is Overfitting with Jean-Francois Puget (Nvidia)
Machine Learning, Artificial Intelligence and Data Science podcast with Jean-Francois Puget
The Information Bottleneck, Hosted by Ravid Shwartz Ziv and Allen Roush
About
NVIDIA researcher Jean-Francois Puget discusses evaluating agent skills and distinguishing real improvements from benchmark overfitting. Drawing on Kaggle competitions, he emphasizes validation that separates development feedback from final evaluation. The episode also considers coding agents, multi-agent software development, model-release claims and his team’s approach to the ARC-AGI competition. Hosted by Ravid Shwartz Ziv and Allen Roush. Watch the full conversation on the publisher’s YouTube channel.
Topics
benchmark overfitting
More from this series
- The Model Found a Way Out - with Florian Brand (Prime Intellect)Jul 27, 2026 · 57 min
- AI Agents and The Golden Age of Asking Questions with Dimitris Papailiopoulos (MSR/UW-Madison)Jul 9, 2026 · 1 h 13 min
- Nathan Lambert: Inside Post-Training and the Open Model FightAug 8, 2026 · 1 h 15 min
- Continual Learning Is the Next Bottleneck | Rohan Anil (Core Automation )Sep 10, 2026 · 1 h 11 min
- Why You Can't Just Rent 1,000 GPUs | Charles Frye (Modal)Sep 1, 2026 · 1 h 3 min
More from The Information Bottleneck
All episodesYuandong Tian on Recursive Self-Improvement
Oct 5, 2026The Information BottleneckChris Manning: Language Is the Real Unlock for Intelligence
Sep 27, 2026The Information BottleneckWhy AI Still Can't See with Andrew Dai
Sep 24, 2026The Information Bottleneck
More with Ravid Shwartz Ziv
All their workCan AI Replace Doctors? Zachary Lipton on the Future of Healthcare and AI
Episode · Sep 21, 2026The Information BottleneckSara Hooker on the End of Static AI
Episode · Sep 16, 2026The Information BottleneckTiny Recursive Models Beat the Giants - Alexia Jolicoeur-Martineau (Microsoft)
Episode · Sep 15, 2026The Information BottleneckWorld Models | John Langford (Microsoft AI Labs)
Episode · Sep 5, 2026The Information Bottleneck
More with Allen Roush
All their workWhich Tabular Model Should You Actually Use? | David Holzmüller (INRIA)
Episode · Sep 3, 2026The Information BottleneckStella Biderman (EleutherAI) - Open Source, AI Safety, and Who We Can Trust
Episode · Aug 28, 2026The Information BottleneckWhy Deep Learning Finally Works on Tables | Frank Hutter (Prior Labs)
Episode · Aug 24, 2026The Information Bottleneck
New episodes in Machine Learning
All episodesPierre-Carl Langlais on Building Models from Data You Can Account For
Jul 23, 2026The Information BottleneckWhy All Models Learn the Same Thing with Phillip Isola (MIT)
Jul 2, 2026The Information BottleneckText Diffusion Models with Brendan O'Donoghue (Google DeepMind)
Aug 14, 2026The Information Bottleneck