AI/ML arXiv cs.AI

When AI Blurs the Boundaries of Contribution: An Empirical Study of Authorship Calibration

An empirical study on 'authorship calibration' explores how heavy AI use leads users to misjudge their actual contribution to AI-generated content.

Other Hacker News

Designing emoji for the way we communicate today

A discussion on the design processes and communication goals behind modern emoji creation.

Tech Business/VC Hacker News

Short sellers notch $8.7B profit as SpaceX shares dip to IPO price

Short sellers made significant profits as SpaceX shares dropped toward the IPO price.

AI/ML arXiv cs.AI

FlashDecoder: Real-Time Latent-to-Pixel Streaming Decoder with Transformers

Introduction of FlashDecoder, a pure-Transformer video decoder that enables real-time, memory-efficient streaming video generation at 1080p.

AI/ML arXiv cs.AI

Show Me How You Reason and I'll Tell You Who You Are: Reasoning Graphs for Robust LLM Authorship Attribution

A new GNN-based approach for LLM authorship attribution that uses reasoning graphs to be more robust against paraphrasing attacks.

Cybersecurity arXiv cs.AI

Random Logit Scaling: Defending Deep Neural Networks Against Black-Box Score-Based Adversarial Example Attacks

Proposed Random Logit Scaling (RLS) as a plug-and-play defense to protect neural networks from black-box score-based adversarial attacks.

AI/ML arXiv cs.AI

Benchmarking Face Recognition without Real Faces

Research demonstrating that high-fidelity synthetic datasets like MorphFace and Vec2Face can replace real faces for benchmarking face recognition models.

AI/ML arXiv cs.AI

A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems

Introduction of DynaBase, a minimal two-parameter architecture for zero-shot reconstruction of dynamical systems, reducing complexity while maintaining performance.

AI/ML arXiv cs.AI

Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control

Development of WA-LQR, a controller that uses mechanistic interpretability and optimal control to improve the robustness of World Action Models (WAMs).

AI/ML arXiv cs.AI

Multi-Axis Max@K Reinforcement Learning for Representative Diversity in Text-to-Image Generation

A new RL objective called multi-axis max@K to improve the representative diversity and fairness of text-to-image generation models.

AI/ML arXiv cs.AI

Latent Trajectory Discrimination for AI-Generated Text Detection

The GTCL framework detects AI-generated text by analyzing the geometric trajectories of latent representations during the autoregressive generation process.

Software Engineering Hacker News

Frame – the first Linux Assembly X server

Announcement of Frame, the first X server written in Linux Assembly, aiming for extreme efficiency and minimal overhead.

Other Hacker News

Frank Lloyd Wright's First Home

A look at the first home designed by architect Frank Lloyd Wright.

Other Hacker News

CO2 overload, detected in human blood, suggests toxic atmosphere within 50 years

Research suggesting that CO2 levels detected in human blood could indicate a toxic atmosphere within 50 years.

Hardware/Chips The Verge

Asus’ top-end 4K QD-OLED gaming monitor is $400 off

A discount offer for the Asus ROG Swift 4K QD-OLED gaming monitor, highlighting its 240Hz refresh rate and G-Sync support.

AI/ML arXiv cs.AI

Interventional Causal Circuits for Safe Robot Action Testing and Failure Recovery

A proposed framework using Joint Probability Trees and Causal Circuits for safe robot action testing and efficient failure recovery in ROS2 environments.

AI/ML arXiv cs.AI

RW-Voice-EQ Bench: A Real World Benchmark for Evaluating Voice AI Systems

Introduction of the RW-Voice-EQ Bench, a comprehensive benchmark for evaluating the acoustic and expressive capabilities of voice AI systems.

AI/ML arXiv cs.AI

Asymmetric Peak-Aware Loss for Peak-Critical Time Series Forecasting

The Asymmetric Peak-Aware Loss (APAL) objective is introduced to improve the prediction of rare demand spikes in time-series forecasting.

AI/ML arXiv cs.AI

Does generative AI supersede supervised XMLC? A Benchmark Study on Automated Subject Indexing with German Scientific Literature

A study comparing supervised XMLC methods with LLM-based generative methods for automated subject indexing of German scientific literature.

AI/ML arXiv cs.AI

Innocuous-Seeming Data, Latent Ideology: Ideological Generalisation in Finetuned LLMs

Research demonstrating 'ideological generalisation', where finetuning LLMs on narrow datasets causes broad, unintended ideological shifts across unrelated domains.