Francois Chaubard
Showing 1–2 of 2 transcripts.
- Y Combinator1h 16m
Kernels and Chips: Cursor, NVIDIA, and Meta Researchers on GPU Performance | YC Paper Club
Stuart Sul, John, Francois Chaubard, Jon Saad-Falcon, Mark Saroufim, Misha Smelyanskiy, Brennan Shacklett
The event synthesizes critical industry shifts toward specialized ASICs and disaggregated architectures that separate training and inference workloads to maximize intelligence per watt. Technical deep dives highlight breakthroughs in multi-GPU kernel optimization and local inference, demonstrating that consumer accelerators can now deliver nearly 90% of frontier model utility while reducing energy costs by 70%. These innovations collectively address systemic bottlenecks in network communication and latency, establishing a roadmap for distributed, cost-efficient AI ecosystems that minimize reliance on centralized cloud resources.
- Y Combinator38 min
Recursion Is The Next Scaling Law In AI
Ankit Gupta, Francois Chaubard
Two 2025 research initiatives, Hierarchical Reasoning Models (HRM) and Tiny Recursive Models (TRM), challenge standard scaling laws by utilizing inference-time recursion to achieve state-of-the-art reasoning with drastically fewer parameters. HRM reaches 27 million parameters using a three-level weighted hierarchy, while the simplified TRM distills the architecture to just 7 million parameters yet achieves 87% accuracy on ARC-Prize benchmarks by treating recurrence as a dynamic latent memory tape. These systems overcome historical RNN limitations through Deep Equilibrium Models and latent recursion, offering a pathway to efficient, deep reasoning that diverges from traditional Chain-of-Thought constraints.