Hardware/Chips The Verge

Lenovo Googlebook leaks reveal a laptop and 2-in-1 tablet

Leaked images reveal upcoming Googlebook laptops and 2-in-1 tablets, intended as successors to Chromebooks.

AI/ML arXiv cs.AI

SAF-OPD: Stable Advantage Fusion for On-Policy Distillation

Introduces SAF, a framework that stabilizes the fusion of RLVR and on-policy distillation in RL to prevent entropy collapse and improve reasoning/coding tasks.

AI/ML arXiv cs.AI

MOSAIC: Masked Outsourcing of Secure AI Computations

Presents MOSAIC, a secure matrix-multiplication masking protocol for outsourcing AI computations to untrusted servers while maintaining privacy and efficiency.

AI/ML arXiv cs.AI

Linear Proposal Operators and Stochastic Search Geometry in SOMA and Differential Evolution

Provides a theoretical analysis of the search geometry in SOMA and Differential Evolution algorithms to improve population-based optimizers.

AI/ML arXiv cs.AI

FBFM: A Training-Free Asynchronous Feedback Mechanism for Flow-Matching in World-Action Models Execution

Proposes Feedback Flow Matching (FBFM), a training-free mechanism for real-time error correction in world-action models for robotics.

AI/ML arXiv cs.AI

Small Is Enough: Per-User Style Rewriting of AI-Edited Text via LoRA Adapters

Presents InMyStyle, a privacy-focused system using small LoRA adapters to rewrite AI-generated text into a user's specific writing style.

AI/ML arXiv cs.AI

When Model Priors Conflict with Visual Evidence: Mitigating Commonsense-Driven Hallucinations by Selective Prior Calibration

Introduces Selective Prior Calibration (SPC) to reduce hallucinations in vision-language models when model priors conflict with visual evidence.

AI/ML arXiv cs.AI

RecHarness: A Bandit-Routed Agentic Harness for Self-Evolving Recommender Systems

Introduces RecHarness, an automated framework for optimizing recommender systems using a bandit router and LLM-generated hypotheses.

AI/ML arXiv cs.AI

TAVI-TEC: An AI-Based Tool for Procedural Planning of Transcatheter Aortic Valve Implantation

Presents TAVI-TEC, an AI-powered framework for automated procedural planning of transcatheter aortic valve implantation using CTA scans.

AI/ML arXiv cs.AI

CalibratedRubric: Task-Adaptive Rubric Banks for Open-Ended LLM Evaluation

Proposes CalibratedRubric, a task-adaptive framework for evaluating open-ended LLM outputs using Bayesian filtering and IRT-based bank assembly.

AI/ML Hacker News

AirLLM 70B inference with single 4GB GPU

AirLLM enables the inference of 70B parameter LLMs on a single 4GB GPU by optimizing memory management.

Other Hacker News

The Abandoned Fish Sauce Terrorizing a Small Canadian Town

A story about abandoned fish sauce causing issues in a small Canadian town.

Tech Business/VC Hacker News

The fading American Dream: Visa uncertainty drives Indian tech workers back home

Indian tech workers are returning home due to increasing visa uncertainty in the United States.

Tech Business/VC TechCrunch

Horizon3 hits $2 billion valuation with $250M Series E as AI threats escalate

Cybersecurity startup Horizon3 raised $250 million, reaching a $2 billion valuation to focus on AI-powered security validation.

Cybersecurity TechCrunch

Samsung bans smart TV apps that share users’ internet connections with strangers

Samsung has banned smart TV apps that facilitate residential proxy networks by sharing users' internet connections.

Other The Verge

Big Walk is like co-op Breath of the Wild

A preview of 'Big Walk', a new co-op multiplayer puzzle exploration game from the creators of Untitled Goose Game.

Other Ars Technica

How headlights got brighter, whiter, and more blinding after dark

An exploration of how automotive headlights have become brighter and more blinding over time.

AI/ML arXiv cs.AI

MBDiff: Multi-view Behavior-aware Diffusion Model for Probabilistic Utility Data Imputation

MBDiff is a multi-view behavior-aware diffusion model designed for probabilistic utility data imputation to improve billing and forecasting.

AI/ML arXiv cs.AI

MoRAE: Flow-Friendly Self-Supervised Latents for Text-to-Motion Generation

MoRAE introduces flow-friendly self-supervised latents to improve the quality and stability of text-to-motion generation.

AI/ML arXiv cs.AI

SERUM: State Extraction and Refinement for User Modeling

SERUM is a framework that extracts interpretable behavioral models from unstructured egocentric video using hierarchical VLM annotation.