Software Engineering Hacker News

Compiler-Assisted Floating-Point Error Analysis and Profiling with FPChecker

Introduction of FPChecker, a tool for compiler-assisted floating-point error analysis and profiling.

Tech Business/VC TechCrunch

Meta, like SpaceX, looks to turn excess AI compute into cash

Meta is exploring the creation of a cloud infrastructure business to monetize its excess AI compute power.

AI/ML arXiv cs.AI

From Search to Synthesis: Training LLMs as Zero-Shot Workflow Generators

MetaFlow is a new framework that trains LLMs to generate zero-shot workflows using supervised fine-tuning and reinforcement learning.

AI/ML arXiv cs.AI

Why Do Few-Step Text Latents Fail When Image Latents Work? Non-Commitment at Sharp Categorical Readouts

Research analyzing why few-step text latents fail compared to image latents, attributing the issue to decoder sharpness and categorical readout.

AI/ML arXiv cs.AI

Hierarchical Global Attention (HGA)

Hierarchical Global Attention (HGA) is introduced as a drop-in replacement for dense causal attention to enable long-context transformers on limited hardware.

Cybersecurity arXiv cs.AI

Understanding and Evaluating Claw-like Agent Security Through a Computer-Systems Lens

SafeClawArena is a new benchmark for evaluating the security of always-on AI agents, exposing significant vulnerabilities in current platforms.

AI/ML arXiv cs.AI

A Single Rewrite Suffices: Empirical Lessons from Production Skill Description Optimization

Empirical study on optimizing AI agent skill descriptions to reduce routing collisions, finding that a single LLM rewrite is highly effective.

Tech Business/VC Hacker News

Sony will no longer produce discs for PlayStation games starting in January 2028

Sony plans to discontinue the production of physical discs for PlayStation games by January 2028, marking a significant shift toward digital distribution.

AI/ML arXiv cs.AI

Unsupervised Thermodynamics of Molecular Diffusion Models: Action-Operator Semantics and Auditable Free-Energy Readout

Researchers introduce a mathematically consistent action-operator framework for diffusion models to estimate free-energy differences in molecular structures.

AI/ML arXiv cs.AI

A Coherence Law for Trainability in Noisy Equivariant Quantum Neural Networks

A new coherence law is proposed to determine the trainability of noisy equivariant quantum neural networks, linking architecture and open-system dynamics.

AI/ML arXiv cs.AI

Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code

An empirical study on Spec-Driven Development (SDD) reveals a trade-off between output determinism and the ability to perform automated hallucination detection in LLM-generated code.

AI/ML arXiv cs.AI

DSIP: A Dynamic Coordination Planner for Signal-Free Intersections using Diffusion-Model-Based Multi-Agent Motion Planning

DSIP is a multi-agent motion planning framework using diffusion models to optimize vehicle trajectories for signal-free intersections.

AI/ML arXiv cs.AI

Modeling Cell-Cycle-Aware Single-Cell Drug Perturbation Responses

scCycleMol is a cell-cycle-aware perturbation prediction framework designed to predict drug responses in single-cell transcriptional data.

AI/ML arXiv cs.AI

LUMOS: A Semantic Operating-System Layer for Accessibility-Grounded AI Agents

LUMOS is a semantic interaction layer that converts OS accessibility metadata into machine-readable blueprints for AI agents, reducing reliance on visual OCR.

AI/ML arXiv cs.AI

BEST-RQ-2: Contextualize-Then-Predict, a Two-Step Approach for Self-Supervised Audio Representations

BEST-RQ-2 introduces a two-step contextualize-then-predict pretraining scheme for self-supervised audio representations using a ViT context encoder.

Cybersecurity arXiv cs.AI

An AI-Based Solution for Secure Service Provisioning in IoT

A framework using Deep Reinforcement Learning and Federated Learning is proposed to secure service provisioning and monitor behavior in IoT environments.

AI/ML arXiv cs.AI

Accelerometry-Derived Digital Biomarkers for Cardiometabolic Risk: A Population-Representative Tabular Benchmark with Uncertainty Quantification

A new population-representative tabular benchmark for cardiometabolic risk is introduced, utilizing TabPFN v2 for prediction and split conformal prediction for uncertainty.

Software Engineering Hacker News

A deep dive into SmallVector:push_back

A technical exploration of the push_back implementation in the SmallVector class, focusing on its internal mechanics.

Tech Business/VC Hacker News

Nintendo has raised its employees base salary by 10%

Nintendo increases the base salary of its employees by 10%.

Other Hacker News

Physical disc production ending in Jan 2028 for new games on PlayStation

PlayStation will end physical disc production for new games starting in January 2028.