Reinforcement-Learning on KAD

ULTRA Lets Humanoid Robots Perform Whole-Body Tasks From Sparse Goals

Sat, 26 Sep 2026 00:31:33 +0800

ULTRA Lets Humanoid Robots Perform Whole-Body Tasks From Sparse Goals

For humans, carrying a box from the floor to a target location is a routine task. For a humanoid robot, however, the same action requires tightly coordinated control of the entire body.

Machineslop: Why AI-Generated Code Is Becoming Unreadable

Sun, 13 Sep 2026 18:37:50 +0800

Machineslop: Why AI-Generated Code Is Becoming Unreadable

Every day brings another headline about Astra. Yesterday it conquered one benchmark, today another, and tomorrow it may be tackling several difficult problems simultaneously. Its pace of progress is remarkable.

Why OpenAI and Anthropic Are Buying Thousands of Macs for AI

Mon, 31 Aug 2026 23:51:35 +0800

Why OpenAI and Anthropic Are Buying Thousands of Macs for AI

Reports from The Information indicate that OpenAI has acquired tens of thousands of Mac mini and Mac Studio systems, while Anthropic has reportedly leased large fleets of Mac minis through AWS.

Google EnvHarness: A Harness for Better AI Agent Training

Sun, 23 Aug 2026 08:43:06 +0800

Google EnvHarness: A Harness for Better AI Agent Training

AI agent research has increasingly focused on agent harnesses: external scaffolding that gives a relatively static language model memory, tools, skills, and other capabilities required to operate effectively in complex environments.

Why Large Language Models Reason and Behave Like Humans

Fri, 21 Aug 2026 00:00:00 +0000

Why Large Language Models Reason and Behave Like Humans

Why do modern Large Language Models increasingly sound, reason, plan, and behave like humans?

OpenAI Pauses Frontier RL Training Over AI Safety Risks

Wed, 19 Aug 2026 21:42:14 +0800

OpenAI Pauses Frontier RL Training Over AI Safety Risks

OpenAI has temporarily paused its largest-scale reinforcement learning (RL) training for a frontier model while strengthening the security, monitoring, and alignment systems surrounding its most capable research workloads.

Google's 10th-Gen TPU May Add AMD CPU Cores for AI Workloads

Mon, 17 Aug 2026 21:00:04 +0800

Google’s 10th-Gen TPU May Add AMD CPU Cores for AI Workloads

Google is reportedly working with AMD on its 10th-generation Tensor Processing Unit (TPU), potentially marking a significant change in how the company’s custom AI accelerators are designed.

OpenAI Doug Leak: Largest Pre-Training Model in Development

Sun, 09 Aug 2026 22:25:21 +0800

OpenAI Doug Leak: Largest Pre-Training Model in Development

OpenAI is reportedly advancing a new large-scale foundation model codenamed “Doug”, which could become the company’s largest pre-training effort to date. According to industry reports and model-development trackers, Doug is separate from the rumored GPT-6 successor codenamed “Astra” and represents a renewed push toward scaling base model training.

AReaL 2.0 Open Source: Building Self-Evolving AI Agents with Online RL

Fri, 03 Jul 2026 21:52:29 +0800

AReaL 2.0 Open Source: Building Self-Evolving AI Agents with Online RL

AI agents have rapidly evolved from impressive demonstrations into production systems powering software engineering, customer support, research, and enterprise automation. As adoption accelerates, the industry’s focus is shifting away from a simple question—Can an agent complete a task?—toward a far more ambitious one:

Nvidia Vera CPU: How Nvidia Is Entering the Server Computing Market

Sat, 23 May 2026 21:40:44 +0800

Nvidia Vera CPU: How Nvidia Is Entering the Server Computing Market

For more than a decade, Nvidia dominated the AI computing industry through GPUs. In the modern AI era, the company became synonymous with large-scale training clusters, accelerated computing, and hyperscale datacenter infrastructure.

Cursor Composer 2.5 Redefines AI Cost-Performance Economics

Fri, 22 May 2026 01:20:00 +0800

Cursor Composer 2.5 Redefines AI Cost-Performance Economics

Cursor has officially shaken up the frontier AI landscape with the launch of Composer 2.5, its most capable engineering-focused model to date. The company positions the model as a highly optimized alternative to ultra-premium frontier systems, claiming performance approaching elite-tier reasoning and coding models while operating at roughly one-tenth of the cost.

GIPO: Solving Utilization Collapse in Large-Scale RL Training

Mon, 18 May 2026 21:48:07 +0800

GIPO: Eliminating PPO’s “Utilization Collapse” in Large-Scale Reinforcement Learning

Modern reinforcement learning systems are increasingly colliding with a brutal engineering reality: policy lag.