AI/ML arXiv cs.AI

Controllable LLM Reasoning via Sparse Autoencoder-Based Steering

Introduces SAE-Steering, a method using Sparse Autoencoders to control reasoning strategies in Large Reasoning Models (LRMs).

AI/ML arXiv cs.AI

JobMatchAI-An Intelligent Job Matching Platform Using Knowledge Graphs, Semantic Search and Explainable AI

JobMatchAI is a production-ready job matching platform using knowledge graphs, semantic search, and the JobSearch-XS benchmark.

AI/ML arXiv cs.AI

DSevolve: Enabling Real-Time Adaptive Scheduling on Dynamic Flexible Job Shop with LLM-Evolved Heuristic Portfolios

DSevolve is a framework for real-time adaptive scheduling in job shops using LLM-evolved heuristic portfolios and a neural selector.

AI/ML arXiv cs.AI

The Possibility of Artificial Intelligence Becoming a Subject and the Alignment Problem

A theoretical exploration of AI alignment, arguing for an autonomy-supporting 'parenting' approach as AGI attains subject status.

AI/ML arXiv cs.AI

Why Does Grounding Hurt Medical VQA? Benchmarking, Diagnosis, and Fine-Tuning of Vision-Language Models

A study on Medical VQA revealing that grounding-by-cropping often degrades performance and proposing a hybrid fine-tuning approach to restore localization.

AI/ML arXiv cs.AI

The Scaling Properties of Implicit Deductive Reasoning in Transformers

Research investigating the scaling properties of implicit deductive reasoning in Transformers using Horn clauses.

AI/ML arXiv cs.AI

AlphaCrafter: Harnessing Multi-Agent Workflows for Cross-Sectional Quantitative Trading

AlphaCrafter is a multi-agent framework for quantitative trading that replaces loose natural-language workflows with structured policy specifications.

AI/ML Hacker News

A pharmacy chain in Vermont implemented AI for efficiency

A pharmacy chain in Vermont is utilizing AI to improve operational efficiency.

Other Hacker News

Angels in Coptic Magic I: Introduction

An introductory piece on 'Angels in Coptic Magic', discussing historical or mystical texts.

AI/ML arXiv cs.AI

MemLens: A Value-Aware Memory Management System with Interactive Analytics for LLM-based Agents

MemLens is a value-aware memory management system for LLM-based agents that uses an interactive dashboard for analytics and memory lifecycle visualization.

Cybersecurity arXiv cs.AI

Does Runtime Topology Context Improve LLM-Generated Kubernetes Security Patches?

KuTIE improves LLM-generated Kubernetes security patches by incorporating live runtime topology context from Istio and Trivy.

AI/ML arXiv cs.AI

Empirical Evaluation of Out-Of-Distribution Performance of Tabular Foundation Models

An empirical evaluation of nine Tabular Foundation Models (TFMs) showing that they degrade under distribution shifts, similar to classical tabular models.

AI/ML arXiv cs.AI

Pictura: Perspective-View Self-Play at Scale for Driving

Pictura is a GPU-accelerated multi-agent driving simulator that enables large-scale self-play training for driving policies directly from perspective views.

Hardware/Chips arXiv cs.AI

MDTransformer: A Hardware-Software Co-Design of Mode-Division Photonic Transformer Accelerator with Inverse-Designed Coherent Crossbar

MDTransformer is a photonic transformer accelerator that uses mode-division optical dataflow to improve energy efficiency and area reduction in Transformer inference.

AI/ML arXiv cs.AI

$\pi\mathbf{R}^2$: Reactive Real-time Flow Policies

$π²$ is a framework that makes action-chunking flow policies reactive and real-time by splitting conditioning into fast and slow channels.

AI/ML arXiv cs.AI

Pass the Baton: Trajectory-Relayed On-Policy Distillation

Relay-OPD is a trajectory-relayed on-policy distillation method that uses teacher intervention to fix failed reasoning paths in student models.

AI/ML arXiv cs.AI

Diffusion Model-based Parameter Estimation in Dynamic Power Systems

A joint conditional diffusion model-based framework for parameter estimation in dynamic power systems, addressing the non-uniqueness problem.

Cybersecurity Hacker News

Flume Water Monitor 915 MHz Security Is Pretty Good

An analysis of the security of the Flume Water Monitor's 915 MHz communication protocol.

Other Hacker News

Starter Homes Are Piling Up While Luxury Homes Fly Off the Market

Discussion on the housing market trends where starter homes are less available while luxury homes are selling quickly.

AI/ML arXiv cs.AI

SAM3D-Guided Object-Centric Representation Alignment for Vision-Language-Action Models

Proposed SAM3D-Guided Object-Centric Representation Alignment framework to improve 3D understanding in Vision-Language-Action (VLA) models for robot manipulation.