Opinions
The Insurance Industry’s Retreat From AI Threatens to Slow Innovation and Adoption
CSIS researcher Gregory C. Allen argues that insurance has become the de facto regulator of AI deployment in the United…
Opinions
Five Bellwethers for AI’s Next Disruptions
The article identifies five influential AI leaders whose work signals coming disruptions in artificial intelligence development as of mid-September 2026,…
Opinions
AI May Reverse Political Polarization. Don’t Get Excited Yet.
Bloomberg Opinion columnist Stephen Mihm explores whether artificial intelligence might help reduce America’s intense political polarization by presenting balanced information…
Engineering
Flattening Every Memory Peak in Long-Context Mixture-of-Experts Training
Researchers identify four unbounded memory peaks in standard MoE training pipelines — expert dispatch, vocabulary projection, gradient checkpointing, and optimizer…
Engineering
SpectralShift: Effective Context Window Extension of Gated DeltaNet via Spectral Reparameterization
The paper studies long-context extension of Gated DeltaNet, a linear-attention architecture, through the spectral properties of its state transition matrix…
Engineering
VC-Attention: Value Smoothing and Softmax Casting for Low-bit Attention
VC-Attention is a training-free low-bit attention framework for video diffusion transformers that tackles both accuracy loss from value outliers and…
Engineering
HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses
HypoEvolve is a multi-agent framework that uses a generational genetic algorithm to coordinate specialized LLM agents that propose, critique, and…
Engineering
LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence
LimiX-2 introduces the Contextual Mechanism Networks (CMNs) paradigm, pretrained with Context-Conditional Masked Modeling, which shifts tabular in-context learning from predicting…
Engineering
Fathom: Per-Query Read Depth for Sparse Decoding over Offloaded KV Caches
Fathom addresses the bottleneck of scanning offloaded KV caches during long agentic sessions by letting each query decide how many…
Engineering
Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents
The paper proposes XConf, a confidence-estimation method that grounds an LLM’s calibrated confidence in a stored record of its own…
Engineering
The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction
Edge0 is a streaming MoE inference engine that serves large mixture-of-experts models from SSD on consumer hardware by predicting each…
Engineering
Agora: Git as Shared Memory for Collective AutoResearch
Agora is a shared-memory system for autonomous research agents that records every hypothesis, result, and verification as an immutable Git…