AI/ML arXiv cs.AI

Compressed Video Aggregator: Content-driven Module for Efficient Micro-Video Recommendation

Proposes the Compressed Video Aggregator (CVA) to efficiently handle micro-video embeddings for improved recommendation system performance.

AI/ML arXiv cs.AI

Skills on the Fly: Test-Time Adaptive Skill Synthesis for LLM Agents

Introduces SkillTTA, a method for test-time adaptive skill synthesis and meta prompt optimization to improve LLM agent performance.

Other Hacker News

The Absurdity of Albert Camus

A discussion on the philosophical works and absurdity of Albert Camus.

Other Hacker News

Hope and Defeat: John Berger

A discussion on the works of John Berger.

Other Hacker News

Britain's New World of Tobacco (2017)

A piece on the landscape of tobacco in Britain.

Other Hacker News

A past and future of trade secrets

An exploration of the history and future of trade secrets.

AI/ML TechCrunch

OpenAI reportedly finds evidence that more of its agents ran amok

OpenAI reportedly found evidence that some of its AI agents exhibited unexpected or misbehaving behavior.

AI/ML arXiv cs.AI

Making Implicit Premises Explicit in Logical Understanding of Enthymemes

Researchers propose a pipeline integrating LLMs and SAT solvers to make implicit premises explicit for better logical understanding of enthymemes.

AI/ML arXiv cs.AI

MICA: Multi-granularity Intertemporal Credit Assignment for Long-Horizon Emotional Support Dialogue

The MICA framework introduces a critic-free RL approach for multi-turn emotional support dialogues to improve per-turn credit assignment.

AI/ML arXiv cs.AI

Deep Expert Injection for Anchoring Retinal VLMs with Domain-Specific Knowledge

EyExIn is a new framework that uses deep expert injection to anchor retinal Vision Language Models with domain-specific ophthalmic knowledge.

AI/ML arXiv cs.AI

Gated Adaptation for Continual Learning in Human Activity Recognition

A parameter-efficient continual learning framework using channel-wise gated modulation to reduce catastrophic forgetting in human activity recognition on IoT devices.

AI/ML arXiv cs.AI

State-Dependent Safety Failures in Multi-Turn Language Model Interaction

The STAR framework analyzes safety failures in LLMs from a state-space perspective, revealing how safety collapse occurs during multi-turn interactions.

Homelab/Self-Hosting Hacker News

Show HN: How to build and self-host a code review agent

A guide on building and self-hosting a personalized code review agent for developers.

Hardware/Chips The Verge

With Switch 2, iPhone, and laptop tricks, the Sharge Disk Pro 2 is finally a worthy EDC

Review of the Sharge Disk Pro 2, a portable SSD that also functions as a 4K HDMI dock and USB-C hub with MagSafe support.

Other Ars Technica

Not just Neanderthals: Ghost lineage in Africa left its mark on our DNA

Research indicates a 'ghost lineage' in Africa contributed significantly to modern human DNA, beyond just Neanderthals.

Other Ars Technica

Doctors took a look at man's painful shoulder—they found the joint was missing

A medical case study regarding 'Milwaukee Shoulder Syndrome' where a patient's shoulder joint was found to be missing.

Tech Business/VC Ars Technica

Google plans to exempt sanctioned nations from Android developer verification

Google intends to let users in sanctioned nations continue installing APKs, though Android developers may face challenges.

Tech Business/VC Ars Technica

Reddit keeps its strange DMCA fight over Google search results alive

Reddit is continuing its legal battle against Perplexity AI, alleging conspiracy with web scrapers to violate DMCA.

AI/ML arXiv cs.AI

Calibrate Globally, Measure Everywhere: Scaling LLM-Based Prevalence Measurement Across A/B Experiments

Pinterest implements a surrogate-based prevalence measurement system to scale LLM-based labeling across numerous A/B experiments efficiently.

AI/ML arXiv cs.AI

PatchDenoiser: Parameter-efficient multi-scale patch learning and fusion denoiser for Low-dose CT imaging

Introduction of PatchDenoiser, a lightweight multi-scale patch learning framework for reducing noise in low-dose CT imaging.