AI/ML arXiv cs.AI

Trajectory-Aware Clinical Risk Prediction via Severity-Grounded Knowledge Graphs and Retrieval-Augmented Generation

TRACER is a framework for clinical risk prediction that integrates severity-grounded knowledge graphs and RAG to improve outcomes in EHR analysis.

AI/ML arXiv cs.AI

Using LLMs for Explainable, Data-Driven Insight Generation from Time Series

A domain-agnostic framework is proposed to generate explainable, data-driven natural language insights from time series forecasts, reducing LLM hallucinations.

AI/ML arXiv cs.AI

Deep Reinforcement Learning to Master the Asymmetric Strategy of Baghchal

This study evaluates various Deep RL algorithms (DQN, PPO, MuZero) on the asymmetric board game Baghchal, finding MuZero to be the most effective.

AI/ML arXiv cs.AI

Operational Hallucination and Safety Drift in AI Agents

The paper identifies 'Safety Drift' and 'Operational Hallucination' in AI agents and proposes an Action-Aware Supervision Layer to enforce architectural reliability.

AI/ML arXiv cs.AI

AlayaWorld: Interactive Long-Horizon World Modeling -- Full Technical Report

AlayaWorld is an open-source interactive video world model capable of generating high-fps video environments from text, images, or video.

AI/ML arXiv cs.AI

Neuro-Symbolic Meta-Policies for Temporal Knowledge-Graph Memory under Partial Observability

A neuro-symbolic meta-policy is introduced for temporal knowledge-graph memory to improve decision-making under partial observability in RDF-based environments.

Software Engineering Hacker News

Original Apollo 11 Guidance Computer source code for command and lunar modules

The original source code for the Apollo 11 Guidance Computer (AGC), used for the first moon landing, is available for review.

Other Hacker News

Ask HN: US Equivalent of Anabin?

A community discussion on Hacker News seeking a US equivalent to the German academic recognition system, Anabin.

AI/ML arXiv cs.AI

Probabilistic Concept-Aware Steering for Trustworthy LLM Inference

Researchers introduce the Probabilistic Concept-Aware Steering (PCS) framework to improve the interpretability and control of LLM inference via concept-driven steering vectors.

AI/ML arXiv cs.AI

FindStatBench: Evaluating Large Language Models on Combinatorial Code Synthesis

FindStatBench is a new execution benchmark designed to evaluate LLMs on combinatorial code synthesis, specifically mapping objects to integers or other objects.

AI/ML arXiv cs.AI

When JSON Is Not Enough: Semantic Reliability of Schema-Constrained LLM Ordering Agents

OrderBench demonstrates that schema-constrained LLM outputs (like JSON) can still contain significant semantic errors, urging the need for domain-specific verification.

AI/ML arXiv cs.AI

ProbSPARQL: Querying Knowledge Graphs with Multi-dimensional, Uncertain Numeric Data

ProbSPARQL is introduced as a SPARQL extension for querying knowledge graphs containing multi-dimensional, uncertain numeric sensor data.

AI/ML arXiv cs.AI

Position: AI/ML Deepfake Research is Misaligned with AI-Generated Non-Consensual Intimate Imagery (AIG-NCII)

A position paper arguing that deepfake research is misaligned with the reality of non-consensual intimate imagery (AIG-NCII), focusing too much on authenticity detection.

AI/ML arXiv cs.AI

MUX: Continuous Reasoning via Multiplexed Tokens

The MUX method proposes using multiplexed tokens in a latent space to achieve higher-bandwidth, more efficient continuous reasoning in LLMs.

AI/ML arXiv cs.AI

State Compression in Two-Agent LLM Relays: A Closed-World Study of Constraint Preservation

A study on state compression in LLM agent relays shows that structured JSON hand-offs provide higher feasibility accuracy than narrative summarization.

AI/ML arXiv cs.AI

Structured Synthetic Reasoning Data for Arithmetic Fine-Tuning of Small Language Models

Research shows that structured synthetic reasoning data can significantly improve arithmetic performance in small language models like Qwen3-0.6B/1.7B on consumer hardware.

Software Engineering Hacker News

Airglow browser lets users modify YouTube, Gmail and Spotify in real time

Airglow is a browser that enables real-time modification of popular web applications like YouTube, Gmail, and Spotify.

AI/ML Hacker News

Kimi K3: second only to Fable 5 on AA-Briefcase

Kimi K3 model performance is reported as being second only to Fable 5 on the AA-Briefcase benchmark.

Software Engineering Hacker News

Ten ways a check passes while the thing it checks is broken

An exploration of ten common failure modes where a check passes despite the system being broken.

AI/ML arXiv cs.AI

Integro-differential equations in angular stabilization of drone motion by distributed feedback control

A research paper proposing angular stabilization of drone motion using distributed feedback control and integro-differential equations.