Hardware/Chips Hackaday

LineShine Is Fastest Supercomputer at Over 2 Exaflops

A discussion on the LineShine supercomputer, which achieves performance exceeding 2 Exaflops.

Hardware/Chips Hackaday

All The Different Lasers, And How Well They Mark 3D Prints

A comprehensive guide on using various lasers to mark 3D printed plastics, featuring experiments from CNC Kitchen.

Hardware/Chips Hackaday

Laser Scanning A Cave With Homebrew Gear

A project demonstrating the use of homebrew laser scanning equipment to map the interior of a cave.

Open Source Hackaday

FLOSS Weekly Episode 872: I’m Not Satoshi

FLOSS Weekly discusses the Craig Wright legal case and features Tristan Sherliker's open-source project, BunTool.

Software Engineering Hacker News

Show HN: Nimic – Pure Python as a systems language with AOT compilation

Nimic is a project that aims to enable pure Python to be used as a systems language through the use of AOT (Ahead-Of-Time) compilation.

AI/ML arXiv cs.AI

Skills for the future software profession: beyond agentic AI!

Researchers discuss the evolving role of software engineers in the agentic AI era, emphasizing the increasing importance of verification and validation skills.

AI/ML arXiv cs.AI

Alternate loss functions and regression models that achieve robustness to outliers by modulating the learning rate

The paper proposes Square Root Loss (SRL) and Smooth Mean Absolute Error (SMAE) to improve regression model robustness against outliers by modulating the learning rate.

AI/ML arXiv cs.AI

MultiMem: Measuring and Mitigating Memorization in Multi-Modal Contrastive Learning

Introduces MultiMem, a new metric for quantifying and mitigating memorization in multi-modal contrastive learning to improve model generalization.

AI/ML arXiv cs.AI

Diffusion Integrated Gradients: Controllable Path Generation for Flexible Feature Attribution

Presents Diffusion Integrated Gradients (DiffIG), a method that uses diffusion models to create flexible, controllable attribution paths for XAI.

AI/ML arXiv cs.AI

On the Position Bias of On-Policy Distillation

Introduces Importance-Weighted On-Policy Distillation (IW-OPD) to solve position bias in RL, leading to faster convergence and better performance on AIME-2025.

Other arXiv cs.AI

AI Fiction in the Wild

A study on the usage patterns of ChatGPT for fiction generation, highlighting the emergence of 'solipsistic reader-writers'.

AI/ML arXiv cs.AI

Polycepta: Object-Centric Appearance Estimation for Multi-Object Tracking

Polycepta is a new object-centric appearance estimation framework for multi-object tracking that outperforms static descriptors and operates at 90.57 Hz.

Hardware/Chips AI News

The math behind the OpenAI Jalapeño chip

OpenAI is collaborating with Broadcom to develop the 'Jalapeño' ASIC chip to reduce dependence on Nvidia and lower infrastructure costs.

Tech Business/VC AI News

Samsung opens ChatGPT Enterprise and Codex access after AI restrictions

Samsung is expanding employee access to ChatGPT Enterprise and Codex globally to enhance technical and non-technical productivity.

Other Hacker News

Wikipedia Workers in Britain set global first by seeking union recognition

Wikipedia workers in Britain are seeking union recognition, marking a global first for the platform's workforce.

AI/ML arXiv cs.AI

Acquisition state behaves as a structured, measurable variable governing lung-nodule AI: kernel-driven measurement instability and noise-driven detection fragility, invisible to DICOM metadata

Researchers identify that reconstruction kernels in medical CT imaging can cause measurement instability and detection fragility in AI models, invisible to standard DICOM metadata.

AI/ML arXiv cs.AI

AgentRivet: an automated system for producing Rivet routines from journal publications

AgentRivet is an automated LLM-based system designed to produce Rivet routines for particle physics analysis from journal publications.

AI/ML arXiv cs.AI

Surprise-Guided MergeSort: Budget-Efficient Human-in-the-Loop Ranking via Adaptive Comparison Scheduling

The Surprise-Guided MergeSort (SGS) framework uses Vision-Language Models to prioritize comparisons for human-in-the-loop ranking, improving efficiency and accuracy.

AI/ML arXiv cs.AI

Lect\=uraAgents: A Multi-Agent Framework for Adaptive Personalized AI-Assisted Learning and Embodied Teaching

LecturaAgents is a multi-agent framework for personalized AI-assisted learning that uses embodied teaching mechanisms to adapt to individual learners.

AI/ML arXiv cs.AI

Robust Dual-Signal Fusion: Hybrid Neuro-Symbolic Gating with Compressed Chain-of-Thought Refinement for Irony Detection in Social Media Texts

The Robust Dual-Signal (RDS) Fusion framework uses a hybrid neuro-symbolic architecture to improve irony detection in social media texts using small-scale LLMs.