AI/ML arXiv cs.AI

Hierarchical Classification via Cascading Feature Elimination: Application to Human Phenotype Ontology-Aligned Facial Phenotyping (FaceMesh2HPO)

FaceMesh2HPO is a framework using hierarchical PointNet-based pipelines to classify facial phenotypes for clinical diagnosis.

AI/ML Hacker News

Mistral's Robostral Navigate: a state of the art robotics navigation model

Mistral introduces Robostral Navigate, a state-of-the-art model designed for robotics navigation.

Software Engineering Hacker News

Cloudflare Meerkat - Globally distributed consensus

Cloudflare introduces Meerkat, a system for achieving globally distributed consensus.

Cybersecurity Hacker News

OpenBSD has a use-after-free allowing local privilege escalation to root

A use-after-free vulnerability in OpenBSD allows for local privilege escalation to root.

Open Source Hacker News

Chatto is now Open Source

Chatto, an AI-powered chat interface, has transitioned to an open-source model.

Other Hacker News

Show HN: Follow London Trains in 3D

A project demonstrating the ability to track London trains in a 3D visualization in real-time.

Other Hacker News

It seems that the age of reading might be a short anomaly in human history

An opinion piece exploring the hypothesis that the era of reading may be a brief anomaly in human history.

Tech Business/VC TechCrunch

Solo GP Ashley Smith announces second $25M fund to back startups in AI, security and more

Ashley Smith announces the closing of a second $25 million fund via Vermilion Cliffs Ventures to support AI and security startups.

Hardware/Chips TechCrunch

Apple to produce Made in America wireless chips with Broadcom

Apple enters a $30 billion+ agreement with Broadcom to produce custom wireless connectivity chips in the United States.

Tech Business/VC TechCrunch

Venus Aerospace raises $90M Series B to build a new kind of rocket engine

Venus Aerospace has raised $90 million in Series B funding to develop a novel type of rocket engine.

Tech Business/VC TechCrunch

Blue Origin reportedly raising $10B at $130B valuation

Blue Origin is reportedly seeking $10 billion in funding at a valuation of $130 billion.

Homelab/Self-Hosting Hacker News

Tiny data centre used to heat public swimming pool

A small-scale data center is being utilized to provide heating for a public swimming pool.

Other Hacker News

US Food and Drug Administration rejects petition to set PFAS limits in food

The US FDA has rejected a petition to establish limits for PFAS chemicals in food products.

Other Hacker News

Japan's Hayabusa2 probe to conduct flyby of Torifune asteroid

Japan's Hayabusa2 space probe is scheduled to perform a flyby of the Torifune asteroid.

Tech Business/VC Ars Technica

Blue Origin, for the first time, is expected to raise private capital

Blue Origin is expected to seek $10 billion in private capital, potentially valuing the company at $130 billion.

AI/ML arXiv cs.AI

Self-Review Reinforcement Learning (SRRL) with Cross-Episode Memory and Policy Distillation

Researchers introduce Self-Review Reinforcement Learning (SRRL), a framework that embeds self-review steps into RL episodes to improve LLM reasoning and learning efficiency.

AI/ML arXiv cs.AI

Most LLM Conformity Needs No Speaker: Measuring the Speaker-Free Floor in Peer-Pressure Benchmarks

Study reveals that LLM conformity—changing answers based on peer responses—often persists even without an explicit speaker, suggesting a bias toward repeated text.

AI/ML arXiv cs.AI

The yes-no bias of large language models reflects answer order and wording, not shifts in moral judgment

Research shows that yes-no bias in LLMs is driven by answer order and wording (lexical pull) rather than shifts in moral judgment.

AI/ML arXiv cs.AI

Prompt Robustness Is Task-Dependent: Comparing Objective and Belief-Style Questions in LLM Evaluation

Evaluation finds that prompt robustness in LLMs is highly task-dependent, varying significantly between objective questions and subjective belief-style questions.

AI/ML arXiv cs.AI

Harnessing Generative Image Models for Training-Free Primitive Shape Abstraction

A new training-free pipeline leverages generative image models to abstract 3D shapes into geometric primitives for robotics and scene understanding.