AI/ML arXiv cs.AI

Beyond the Hard Budget: Sparsity Regularizers for More Interpretable Top-k Sparse Autoencoders

Research proposing new sparsity regularizers for Top-k Sparse Autoencoders to improve the interpretability of vision foundation models.

AI/ML arXiv cs.AI

Empowering GUI Agents via Autonomous Experience Exploration and Hindsight Experience Utilization for Task Planning

The PEEU method enhances GUI agents' planning abilities in small MLLMs through autonomous exploration and hindsight experience utilization.

Tech Business/VC Hacker News

Hightouch (YC S19) Is Hiring

Hightouch, a Y Combinator startup, is currently hiring for various positions.

Other Hacker News

Pre-Modern Armies for Worldbuilders, Part III: Paying for It

A detailed guide on the economic aspects of paying for pre-modern armies for worldbuilders.

Other Hacker News

Long Wave radio era set to end with Droitwich switch-off

The era of long wave radio is coming to an end with the shutdown of the Droitwich transmitter.

Other Ars Technica

Russian citizens told "switch to Android" after Apple blocks key Russian apps

The Russian government has urged citizens to switch to Android after Apple blocked key Russian apps on the App Store.

Tech Business/VC Ars Technica

NYT slams Microsoft for building copyright-infringing supercomputer for OpenAI

The New York Times is pursuing copyright claims against Microsoft and OpenAI regarding the use of a supercomputer for training AI models.

AI/ML arXiv cs.AI

Learning to Fold: prizewinning solution at LeHome Challenge 2026 (1st place online, 2nd offline)

A prize-winning solution for bimanual garment folding using a vision-language-action (VLA) policy improved with reinforcement learning.

AI/ML arXiv cs.AI

Automating Potential-based Reward Shaping with Vision Language Model Guidance

Introduces VLM-PBRS, a framework that uses Vision Language Models to automate potential-based reward shaping in reinforcement learning.

AI/ML arXiv cs.AI

CARVE: Content-Aware Recurrent with Value Efficiency for Chunk-Parallel Linear Attention

Introduces CARVE, a new recurrent architecture that improves efficiency and performance over GDN-2 by erasing only on the key axis.

AI/ML arXiv cs.AI

Bridging Talk and Thought: Understanding Dialogue Dynamics Across Collaborative Problem-Solving Contexts

A conceptual framework for analyzing dialogue dynamics in human-AI and multi-agent collaborative problem-solving.

AI/ML arXiv cs.AI

From Celebrities to Anyone: Characterizing AI Nudification Content, Technology, and Community Dynamics on 4chan

A study on the proliferation of AI-generated non-consensual sexually explicit imagery, highlighting a shift toward targeting non-celebrities.

Other Hacker News

The Art of Kite Flying (1430–1929)

An exploration of the history and art of kite flying from 1430 to 1929.

Software Engineering Hacker News

Slisp: Simple Lisp compiler (Linux/amd64)

Slisp is a simple Lisp compiler targeting Linux/amd64.

AI/ML arXiv cs.AI

Parametric Open Source Games

Introduces parametric open-source games as a continuous analogue of program equilibria in game theory, showing how open-source coupling can steer optimization toward cooperation.

AI/ML arXiv cs.AI

Beyond Global Divergences: A Local-Mass Perspective on Bayesian Inference

Proposes a local-mass perspective on Bayesian inference using Mass Index and regularised extended KL (RE-KL) to better characterize local distributional discrepancy.

Cybersecurity arXiv cs.AI

Inherited Circuits, Learned Semantics: How Fine-Tuning Creates Evasion Vulnerabilities Invisible to Standard Evaluation

Reveals that security fine-tuning in LLMs can create evasion vulnerabilities by specializing inherited circuits into brittle indicator rules.

AI/ML arXiv cs.AI

Data-Free Reservoir Features for Efficient Long-Horizon Cold-Start Continual Learning

Presents CIRCLE, a class-incremental classifier using fixed reservoir features that eliminates the need for backbone backpropagation in cold-start continual learning.

AI/ML arXiv cs.AI

Application of LLMs to Threat Assessment of Foreign Peacekeeping Missions

Describes a workflow using LLMs and OSINT to extract threats for foreign peacekeeping missions, showing high agreement with human analysts.

AI/ML arXiv cs.AI

Heavy-Ball Q-Learning with Residual Weighting Correction

Proposes a heavy-ball Q-learning method with residual weighting correction to accelerate convergence in reinforcement learning.