Tag
Brendan Hansknecht from Modular explains that their high performance numbers stem from treating performance as a full-stack problem, rather than relying on piecemeal components in production.
ModCon2026, a developer conference by Modular, focused on discussions about AI infrastructure, the economics of training and inference, and the impact of open models, with takeaways highlighted by Dave Munichiello.
This paper introduces scaling participation, a new paradigm for building modular AI systems through contributions from diverse stakeholders, where small models collaborate to outperform monolithic LLMs by up to 15.4% across various tasks, demonstrating emergent capabilities and improved diversity benefits.
Researchers from Amazon AGI introduce Cartridges at Scale (CAS), a training framework that distills document collections into modular, reusable KV caches, enabling scalable multi-cartridge learning over collections exceeding one million tokens. CAS improves over monolithic cartridge baselines by 10–31 points and matches or exceeds conventional RAG accuracy while consuming 3–4× fewer prompt tokens.
Proposes PDRNN, a modular hybrid AI-assisted pedestrian dead reckoning system that combines a recurrent neural network with separate ML models for orientation, velocity, and distance estimation, with optional radio-based stabilization. Experiments on dynamic sports movement data show superior accuracy and precision compared to classic and ML-based methods.