Tag
A preprint finds that large language models spontaneously develop specialized modular brain regions for language, math, physics, and social reasoning, similar to the human brain, suggesting convergence in intelligent system design.
The article argues that open-weight AI models are catching up to closed ones not via distillation but due to the modularisation of the AI stack—stable interfaces (Transformer architecture, OpenAI-compatible APIs, agentic harnesses) allow innovations to diffuse rapidly across the ecosystem, shrinking the capability gap while keeping a massive price advantage, potentially leading to a commoditisation of frontier AI.
ModTGCN is a modularity-aware graph neural network that jointly optimizes cross-entropy and a modularity-based auxiliary objective to improve text classification by leveraging global community structure in document graphs, achieving consistent gains on five benchmarks.
This paper identifies a capacity-induced failure mode in physics-informed neural networks (PINNs) where overparameterized networks develop functional modularity that hinders convergence, and proposes Modular-Sparsity Synchronization (ModSync), a framework that penalizes task-exclusive connections to maintain cross-objective interaction and achieve state-of-the-art accuracy.
This article draws parallels between biological evolution and technological evolution, explaining how modularity and sexual reproduction allow populations to increase the rate of information acquisition. Simulations demonstrate that mixing genetic material accelerates the spread of beneficial mutations, analogous to how technologies build on existing components.
LatentSkill converts textual skills into LoRA adapters stored in weight space, reducing context overhead while maintaining modularity and composability for LLM agents, achieving significant improvements on ALFWorld and Search-QA benchmarks.
This paper proposes principled approaches for designing and optimizing practical agentic LLM systems, introducing a framework with pseudo-tools and fixed workflows to improve modularity, cost-efficiency, and accuracy across diverse tasks.
Allen AI releases EMO, a mixture-of-experts model where modular structure emerges naturally from data, enabling use of just 12.5% of experts for a task while maintaining near full-model performance.