Infrastructure & Hardware B

This arXiv paper tests whether uncertainty estimation (UE) gives clinical vision-language models a reliable trust-or-escalate signal. Benchmarking 8 methods across 12 VLMs on clinical visual question-answering, the authors find UE quality is not intrinsic to the method but tracks model accuracy—degrading exactly where performance is weakest and reliability most needed. Under perturbations that hide the correct option, accuracy collapses while uncertainty barely changes.

This arXiv study tests whether LLM agents can analyze ubiquitous time series data used in finance, healthcare, and environmental monitoring. Comparing three approaches—raw numerical data, the LLM as a coding agent, and a combination—the authors find that agents with code access can outperform models processing raw data by up to 10%, though even the best agent still answers roughly 22-34% incorrectly.

GPU kernel optimization assumes correctness and targets execution efficiency. The authors present daVinci-kernel, an RL framework coupling skill discovery and exploitation via a dynamically evolving skill library. Three agents share one LLM backbone: a Selection Agent retrieving techniques via BM25 and LLM reranking, a Policy Agent generating CUDA/Triton kernels, and a Summary Agent distilling rollouts into reusable skills. Skills are added only after execution verification confirms speedups.

A technical note exploring how to map columns in arbitrary SQLite query results back to their originating table.column, so that Datasette could render queries with richer, source-aware metadata.

Simon Willison updated his browser tool for OpenAI's WebRTC realtime audio API. It now supports the newer realtime voice model touting GPT-5-class reasoning, and lets users paste document text as context for spoken conversations about it.

NVIDIA reports leading agentic coding performance on the first benchmark dedicated to agentic AI, per its developer blog. The result highlights its inference stack and GPU infrastructure as a platform for autonomous coding agents.

AdaSR moves beyond the read-then-think paradigm by letting reasoning models reason incrementally as input streams in. It uses a hierarchical relative policy optimization scheme to train streaming reasoning.

As AI coupled to proof assistants generates formal mathematics at scale, a gap opens between what a checker verifies and what mathematicians value. Through the lens of language generation in the limit, the paper argues that producing trivial, peripheral statements is provably necessary to generate valuable mathematics.

The paper examines whether the Compressed Computation toy model is an instance of computation in superposition. It argues, based on analysis, that it probably is not.

LLMs serve as execution engines for agentic systems yet still consume context through a sequential text interface, mismatching modern structured workflows with independent parallel branches. The paper explores synthesizing such parallel branches directly in latent space.

Through computational analysis, this paper argues that the three movements of Beethoven's Moonlight Sonata (Op. 27 No. 2) instantiate three distinct machine learning architectures by structural correspondence rather than mere analogy.

The study presents a statistical and machine learning framework characterizing a hydrogen-based multi-energy system. It targets operational threshold detection and deployable dispatch controller development.

Post-training INT8 quantization of diffusion transformers is often slower than FP8/NF4 on consumer Ampere GPUs. The paper presents a fused INT8 GEMM kernel for Ideogram 4.0 that realizes native INT8 speedups.

Clustering is widely used but its outputs are hard to interpret and audit. Cluster LOCO provides feature-importance scores to explain what distinguishes each cluster.

Applying GRPO to GUI grounding samples rollouts from a single screenshot, so groups often turn all-failure or all-success and yield weak signal. VISTA introduces view-consistent, self-verified training to stabilize GUI grounding.

Emulators cheaply reproduce regional climate models' downscaling, linking global-model predictors to high-resolution fields. The paper assesses the added value of diffusion-based generative machine learning for regional climate model emulation.

AI evaluations are widely used to track progress, but inconsistencies across evaluators hinder analysis and comparison. The paper proposes a unifying schema and a community repository, Every Eval Ever, for AI evaluation results.

NVIDIA's developer blog explains how to deploy MiniMax M3 on NVIDIA accelerated infrastructure for long-context reasoning and agentic workflows, addressing fragmented enterprise AI pipelines spanning text, vision, and other modalities.

Fine-tuning vision-language models to emit dense coordinate lists improves grounding but alters how they serialize, repeat, and terminate structured output. The paper shows this induces a controllable interference surface in VLMs.

A growing line of work equips LLM agents with graph neural networks as callable tools. The paper finds that agents defer blindly to these GNN tools, and that stronger backbones tend to defer even more.