Author

Machine Heart

Professional AI media and industry service platform

1.2k

Articles

Likes

8.6k

Views

Comments

Latest from Machine Heart

100 recent articles max

Machine Heart

Oct 19, 2026 · Artificial Intelligence

LoopVL: Recurrent Vision-Language Model Reveals 'Visual Aha Moment'

Researchers from Baidu, Renmin University, Tsinghua, and others introduce LoopVL, a recurrent vision-language model that reuses a 1B-parameter language backbone across 128 layer calls, achieving competitive performance with 260x less training data and exhibiting a 'Visual Aha Moment' where attention sharply refocuses across loops.

Attention MechanismLoop TransformerLoopVL

0 likes · 15 min read

LoopVL: Recurrent Vision-Language Model Reveals 'Visual Aha Moment'

Machine Heart

Oct 19, 2026 · Artificial Intelligence

From Seeing to Acting: Smart Glasses as First-Person Intelligence Platforms

Zhejiang University's APRIL Lab publishes a comprehensive survey establishing a systematic framework for smart glasses as first-person intelligence platforms, defining eight hardware capability axes, seven fundamental capabilities, an L0-L5 maturity model, nine application scenarios, and a nine-dimensional deployment framework with evaluation protocols.

Embodied AIegocentric visionfirst-person vision

1 likes · 8 min read

From Seeing to Acting: Smart Glasses as First-Person Intelligence Platforms

Machine Heart

Oct 19, 2026 · Artificial Intelligence

From One-Model-Fits-All to Personalized AI: How Tongyan Lingzhi Moves Personalization into Model Parameters

Tongyan Lingzhi builds a continuous self-improving personal AI system using a shared base model plus lightweight Personal LoRA parameters, driven by a Recursive Self-Improvement (RSI) loop that learns from real user interactions, supported by training-inference parallelism (TIP) and Paged LoRA for efficient scaling.

AI infrastructureContinual LearningLoRA

0 likes · 15 min read

From One-Model-Fits-All to Personalized AI: How Tongyan Lingzhi Moves Personalization into Model Parameters

Machine Heart

Oct 19, 2026 · Artificial Intelligence

DynStream: First Pose-Free Streaming 4D Gaussian Reconstruction for Large-Scale Dynamic Scenes

Peking University, VGI Labs, and UC Merced introduce DynStream, a feed-forward framework that reconstructs dynamic 3D scenes from pose-free video streams by chunking input, predicting relative poses and scale factors for global alignment, and separating static and dynamic Gaussians for continuous rendering.

4D reconstructionGaussian splattingcomputer vision

0 likes · 9 min read

DynStream: First Pose-Free Streaming 4D Gaussian Reconstruction for Large-Scale Dynamic Scenes

Machine Heart

Oct 18, 2026 · Artificial Intelligence

DR-Rubric: Deep Research as Rubric for Open-Ended Reinforcement Learning

Xiaohongshu and Fudan University propose DR-Rubric, a framework where models perform deep research to gather evidence and synthesize atomic constraints as rubrics for reinforcement learning rewards, achieving competitive results on six benchmarks with only 1,000 training samples while revealing non-monotonic bootstrap dynamics and reward polarization as a stopping signal.

BootstrapDR-RubricDeep Research

0 likes · 21 min read

DR-Rubric: Deep Research as Rubric for Open-Ended Reinforcement Learning

Machine Heart

Oct 18, 2026 · Artificial Intelligence

SPADER: Step-wise GRPO with Diversity-Aware Rewards Boosts Long-Tail Answer Discovery in Multi-Answer QA

SPADER introduces a step-wise GRPO framework with diversity-aware exploration rewards to improve long-tail answer coverage in multi-answer question answering, achieving state-of-the-art results across four benchmarks and two base models by refining credit assignment and incentivizing rare entity discovery.

Diversity-Aware RewardsEMNLP 2026GRPO

0 likes · 16 min read

SPADER: Step-wise GRPO with Diversity-Aware Rewards Boosts Long-Tail Answer Discovery in Multi-Answer QA

Machine Heart

Oct 18, 2026 · Artificial Intelligence

Why Source Code Is the Untapped Resource for Coding Agent Training

This article analyzes how recent research redefines source code's role in training coding agents, moving beyond mere training corpus to serve as executable task environments, verification oracles, and skill repositories, citing works like CodeMidas, R2E-Gym, and SWE-smith that leverage implemented functionality for task generation and validation.

AI TrainingCodeMidasCoding Agents

0 likes · 6 min read

Why Source Code Is the Untapped Resource for Coding Agent Training

Machine Heart

Oct 18, 2026 · Artificial Intelligence

Why Meta's Muse Beat OpenClaw: Inside Alexandr Wang's AI Agent Playbook

Meta's AI agent Muse reached 5M US downloads in 22 days by focusing on personal use cases over coding, rigorously engineering reliability across 100+ behaviors, keeping the team under 200 people, leveraging meme-driven marketing, and building user trust through a gradual delegation ladder.

AI AgentAlexandr WangMeta

0 likes · 16 min read

Why Meta's Muse Beat OpenClaw: Inside Alexandr Wang's AI Agent Playbook

Machine Heart

Oct 18, 2026 · Artificial Intelligence

Elad Hazan's 20-Year OCO Retrospective: Regret Minimization, AdaGrad, and AI Frameworks

Princeton professor Elad Hazan recounts 20 years of Online Convex Optimization research, from inventing Online Newton Step and co-creating AdaGrad to developing spectral filtering for dynamical systems and Flash STU architectures, arguing that as AI excels at theorem-proving, human value lies in choosing problems and inventing frameworks.

AdaGradMachine LearningOnline Convex Optimization

0 likes · 18 min read

Elad Hazan's 20-Year OCO Retrospective: Regret Minimization, AdaGrad, and AI Frameworks

Machine Heart

Oct 17, 2026 · Artificial Intelligence

EgoWild2Dex: Robots Master Complex Tasks from 538 Hours of 'Dirty' Human Egocentric Data

Hong Kong University's MMLab introduces EgoWild2Dex, a framework that trains robots on 538.9 hours of uncurated human egocentric video using GeoFormer for view alignment and progressive human-robot co-training, enabling complex bimanual tasks like package opening and drink making with under 1 hour of real robot data and 96.7% success rate.

EgoWild datasetEgoWild2DexGeoFormer

0 likes · 13 min read

EgoWild2Dex: Robots Master Complex Tasks from 538 Hours of 'Dirty' Human Egocentric Data