Can LLM Agents Design Chips From Higher-Level Abstractions?
A new preprint asks whether large language model agents can outperform RTL-level approaches by designing chips from higher-level abstractions.
Research Digest
Preprints and papers, rewritten in plain English.
LLM Research Digest: Efficiency Gains and Reliability Gaps
New papers show rapid progress in making LLMs cheaper to run, while others expose persistent fragility in constrained and long-horizon settings.
AI agent authorization: five papers, one gap, different fixes
New arXiv papers agree that static credentials and model rationales cannot govern autonomous agents, but they diverge on where to intervene.
Skeleton-guided world models turn human videos into robot training data
Seven arXiv papers demonstrate how human demonstrations, skeletons, and world models can reduce reliance on costly robot-collected training data.
Three new attacks on the KV cache bottleneck
New papers tackle LLM memory pressure from three directions: hardware interconnect, entropy coding, and compression for input-heavy agents.
New Preprints Tackle Vision-Language Navigation in Continuous, Onboard Settings
Three arXiv preprints examine how VLN-CE agents handle long-horizon instructions in unknown spaces, from zero-shot language-model reasoning to fully onboard aerial robots.
Evaluating LLM Code: Translation Benchmarks and Metric Validity
Three new papers highlight the need for more rigorous evaluation of LLM code generation and the metrics used to judge code quality.
Three New Papers Tackle the Sim-to-Real Gap in Autonomous Driving
Recent arXiv preprints propose different fixes—mid-level representations, real-traffic simulation, and render-free teacher fine-tuning—for the persistent mismatch between training environments and real roads.
LLM Groups Can Overstate Consensus and Settle on Wrong Answers
Three arXiv preprints examine when LLM collectives reach reliable consensus—and why interaction design matters as much as reasoning.
New Locomotion Research Tackles Slopes, Terrain, and Robot Design
Three preprints tackle legged locomotion beyond flat floors, from pitched roofs to terrain-aware tracking and co-designed robot bodies.
Three New Papers Rethink AI Privacy Beyond Training-Time Defenses
Recent research shifts privacy focus to retrieval deletion, semantic sanitization metrics, and the protective effect of missing data.
Three Papers Rethink 3D Scenes for Embodied Agents
New work moves indoor 3D scene research from static visual maps toward queryable, activity-aware, and functionally generated environments.
Odometry research tackles calibration, tuning, and radar noise
Three new papers push odometry beyond fixed camera settings, manual tuning, and sparse radar returns.
Diffusion Language Models Get a Closer Look: Training, Parallelism, and Watermarks
Three new papers examine the promise and practical challenges of diffusion-based large language models, from efficient long-context training to provenance.
Neural PDE Solvers: New Work on Training, Reuse, and Spatial Adaptivity
Recent preprints propose frameworks to make physics-informed neural networks and neural operators faster, more reusable, and better at handling spatially varying equations.
New Studies Probe How LLMs Store, Forget, and Distort Knowledge
Three papers examine the mechanics of knowledge in large language models, from tracing its origins to removing it and warning against its misuse.
Open-Vocabulary 3D Perception Extends Beyond Fixed Category Sets
Three papers push 3D object detection and segmentation beyond closed category lists, but they target different tasks and sensor inputs.
Historical OCR: VLMs Face Cost and Hallucination Hurdles
Three arXiv papers examine whether vision-language models can replace traditional OCR for historical documents, with cost and hallucination emerging as key barriers.