AI/ML arXiv cs.AI

CompDiff: Hierarchical Compositional Diffusion for Fair and Zero-Shot Intersectional Medical Image Generation

CompDiff is a hierarchical compositional diffusion framework designed to generate fair and zero-shot intersectional medical images.

AI/ML arXiv cs.AI

Effective Strategies for Asynchronous Software Engineering Agents

The CAID paradigm improves multi-agent collaboration for software engineering by using centralized delegation, asynchronous execution, and isolated workspaces.

AI/ML arXiv cs.AI

Object Search in Partially-Known Environments via LLM-informed Model-based Planning and Prompt Selection

A new LLM-informed model-based planning framework enhances object search in partially-known environments using optimized prompt selection.

Software Engineering Hacker News

Interview with Mitchell Hashimoto about Ghostty and Zig

An interview with Mitchell Hashimoto discussing the development of the Ghostty terminal emulator and the Zig programming language.

AI/ML TechCrunch

OpenAI launches its new family of models with GPT-5.6

OpenAI announces a new family of models, GPT-5.6, with reported improvements in general capabilities and cybersecurity.

Tech Business/VC TechCrunch

An AI agent startup just let its agent run its $100 million fundraise

AI agent startup Lyzr claims to have used its own AI agent to facilitate its $100 million funding round.

AI/ML arXiv cs.AI

FDRMFL: Multimodal Federated Feature Extraction Model Based on Information Maximization and Contrastive Learning

Researchers propose FDRMFL, a multimodal federated feature extraction framework designed for regression under non-IID data distributions.

AI/ML arXiv cs.AI

HiMoE-VLA: Hierarchical Mixture-of-Experts for Generalist Vision-Language-Action Policies

HiMoE-VLA is introduced as a Hierarchical Mixture-of-Experts framework for generalist vision-language-action policies to handle diverse robot embodiments.

Hardware/Chips arXiv cs.AI

HiDVFS: Hierarchical Multi-Agent DVFS for Real-Time OpenMP DAG Workloads

HiDVFS is presented as a hierarchical multi-agent DVFS scheduler for improving energy efficiency and meeting deadlines in OpenMP DAG workloads on embedded systems.

AI/ML arXiv cs.AI

ButterflyMoE: Compression-Scalable Ternary Experts via Structured Butterfly Orbits

ButterflyMoE introduces a memory-efficient MoE architecture using structured butterfly orbits to reduce expert memory costs from O(d^2) to O(d log d).

AI/ML arXiv cs.AI

Spatiotemporal Semantic V2X Framework for Cooperative Collision Prediction

A semantic V2X framework utilizing V-JEPA embeddings is proposed to enable cooperative collision prediction in intelligent transportation systems while reducing bandwidth.

AI/ML arXiv cs.AI

Can We Really Learn One Representation to Optimize All Rewards?

Theoretical analysis of forward-backward representation learning in RL leads to a new variant that significantly reduces error and improves zero-shot performance.

AI/ML arXiv cs.AI

Named-Entity Recognition in the Crime Domain (CrimeNER): Case Study and Dataset

The CrimeNER paper introduces a specialized dataset (CrimeNER-db) and a case study for Named-Entity Recognition in the crime domain.

Hardware/Chips Hacker News

Meta reuses old RAM in new servers with custom bridge chip

Meta is implementing a custom bridge chip to enable the reuse of older RAM modules in its latest generation of servers.

Software Engineering Hacker News

Show HN: Rubiks Cube Solver

A developer has shared a custom software implementation for solving Rubik's Cubes on Hacker News.

Tech Business/VC TechCrunch

OpenAI is shutting down Atlas, but its AI browser ambitions are still growing

OpenAI is shutting down its AI-powered browser, Atlas, while shifting some of its agentic capabilities to its desktop app and Chrome extension.

Tech Business/VC TechCrunch

Elon Musk praises Mythos/Fable, promises not to ‘cut off’ Anthropic

Elon Musk has expressed support for Mythos/Fable and assured Anthropic that it can trust his infrastructure to host its AI models.

Tech Business/VC TechCrunch

Can AI answer the $3 trillion question?

An exploration of the return on investment (ROI) debate surrounding the massive financial expenditures in the AI industry.

Cybersecurity arXiv cs.AI

NonTextual Target Attack

Researchers propose NonTextual Target Attack (NTA), a gradient-based jailbreak method for LLMs that maximizes unsafety probability without requiring fixed target responses.

AI/ML arXiv cs.AI

Rapidly Learning Soft Robot Control via Implicit Time-Stepping

A new approach using implicit time-stepping and the DisMech simulator enables rapid policy learning for soft robots, significantly outperforming previous frameworks.