Engineering
OpenGCUA Open-Sources a Full Stack for Generalist Computer-Use Agents
Researchers have released OpenGCUA, an open-source project providing a full stack for generalist computer-use agents that observe, act, and organize…
Engineering
Embeddings Are Overpriced? A Frozen 16 MB Vocabulary for Small and Ternary Language Models
A new paper proposes replacing a small language model’s learned embedding table, often its largest set of parameters, with a…
Engineering
A Conditional No-Hair Theorem Explains How Transformers Forget Their Prompt
A paper posted to arXiv on October 16, 2026 by Shuyang Xiang proves a “conditional no-hair theorem” for autoregressive transformers,…
Engineering
TokenRouter: Efficient Serving System for Token-Level LLM Routing
TokenRouter is a serving system designed for token-level routing across multiple large language models within a single inference pipeline. Rather…
Engineering
REMORY: Learning Residual Memory for Context Compaction
REMORY is a neural memory mechanism designed to address information loss that occurs when long-running AI agents compress their interaction…
Engineering
Scaling to Tens of Thousands of Test-Time Iterations with Loop-Native Attention Residuals
Researchers introduced InfiLoop, a residual-connection design for looped transformer models intended to prevent reasoning accuracy from degrading as more test-time…
Engineering
AgentGarten: Code Worlds for Evolving Agents
Researchers have introduced AgentGarten, a framework that pairs simulators and game engines with a shared neural renderer to build real-time…
Engineering
Impactful Scheduling for GPU Clusters
The Allen Institute for AI’s (Ai2) infrastructure team replaced the priority-based scheduler on its GPU clusters, which serve about 150…
Engineering
Clinician Use of Language Models Diverges From How the Models Are Evaluated
Researchers analyzed 127,833 real-world clinical queries from 6,342 physicians, advanced practice providers, and nurses across 35 specialties during an eight-month…
Engineering
TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models
Researchers introduce TRACE, an FP4 quantization framework that improves reinforcement learning training efficiency for Mixture-of-Experts language models. The method applies…
Engineering
UNREAL: Unifying Retrieval and Long-Context with a Single Model
UNREAL is a framework that unifies evidence selection for corpus retrieval and long-context inference through a single model-internal mechanism, adding…
Engineering
NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale
NeMo-DCR is a delta-compression technique for synchronizing trillion-parameter models across distributed reinforcement learning systems. Observing that only about 1% of…