Press ESC to close Press ⌘K or Ctrl+K to open

用行动赢得尊重

Alan Hou

Notes on coding, AI, and technology.

记录编程、人工智能与技术探索。

Recent Posts / 近期文章

View all posts

See Every Block Once: Variance-Reduced Post-Training for Masked Diffusion Language Models

每个块只看一次:掩码扩散语言模型的低方差强化学习后训练

Classical Energies Learn to Reason 1,000x Faster than Looped Models

经典能量模型推理学习速度超越循环模型千倍:重构 AI 逻辑推理范式

Two Mathematical Models of Knowledge Distillation: Why Teachers Accelerate Optimization Without Changing Generalization

知识蒸馏的双重数学模型:为什么教师模型加速优化却不改变泛化上限

Back to Pixels: Lifting Latent Video Diffusion into Pixel Space

重返像素空间:摆脱 3D VAE 桎梏的原生像素级视频扩散模型 PixVid

Unbundling Multimodal Vision Encoding: Why Pretrained Priors Are Optional but Dedicated In-Fusion Capacity Is Key

解构多模态视觉编码:为什么预训练先验可有可无,而早期融合内专用算力才是关键

Is Your Robot's Memory Just a Clock? Temporal Shortcuts in Memory-Augmented VLA Policies

机器人的记忆难道只是一块时钟?揭秘具身 VLA 大模型中的时间捷径与虚假进度

A Theory of Policy Optimization in Post-Training: Coverage Collapse and Reference Resets

后训练策略优化的理论基石:覆盖度崩溃与参考策略重置机制

BiVerify: Closing the Loop Between Backward and Forward Verification for Mathematical Self-Correction

BiVerify:闭环双向验证框架——攻克大模型数学自我纠错的推理瓶颈

GenRM-Flow: Video Generators Inherently Function as Process-Aware Reward Models in Flow Matching

GenRM-Flow:流匹配视频生成器本身就是绝佳的过程感知奖励模型

Series / 系列文章