X

NASA shared a James Webb Space Telescope image of the Cat's Head Nebula, named for its feline-like shape resembling a cosmic cat sitting with its back to the viewer.

Inspired by Karpathy’s HackerTube, the author used Teamily AI’s multi-agent workflow to build “AnliBili,” a website that curates 1,132 high-quality Bilibili videos repeatedly recommended across nearly 10 million discussions spanning 16 years on Zhihu and Xiaohongshu, helping users cut through information noise.

The official companion code repository for O'Reilly's bestselling book 'Hands-On Large Language Models', featuring nearly 300 illustrations and Jupyter Notebooks that run directly on Google Colab, covering full-stack practices including Transformers, prompt engineering, RAG, multimodal models, and fine-tuning. Has earned 30K+ Stars.

Paper proposes Skill++, which rewrites only the name and description of incoming agent skills to prevent routing interference in growing skill libraries, raising full-skill selection from 36.4% to 76.3% across 731 skills, three benchmarks, five routers, and the Codex CLI harness.

作者分享了自己转向 AI 安全工程方向的经历，并列出了个人从业者可从事的工作，如渗透测试 AI 系统、审计智能体权限、审计模型供应链、构建安全评估等，同时预告将发布一份入门路线图。

An empirical arXiv study examines /retro, a skill that lets coding agents review completed sessions and propose harness changes, finding that while many proposals are generated, only some reduce repeated work on later tasks — showing harness self-evolution needs mechanisms to verify and revise changes.

A user shares a tested tutorial: subscribing to Gemini AI Ultra 20X at around 410 yuan per month through the Apple payment channel on a Turkish-region Apple account, claiming the value for money is enough to replace a GPT Pro subscription.

Matt Pocock 试验一种组合架构：本地长期运行的 orchestrator 线程与软件工厂（沙盒化实现者、探索任务等）相连，通过 CLI 管理线程、复盘和 cron 任务，让人类用自然语言交互。

WPBeginner's October roundup highlights WordPress plugins gaining AI agent capabilities (OptinMonster MCP, WPVibe.ai surpassing 100k users and joining Claude's Connectors Directory), the WP Crontrol 2.0 handoff, and critical PHP vulnerabilities disclosed on October 21.

Someone has released an open-source tool called 'gpu-unlock', claiming it can remove NVIDIA's software restrictions on GeForce graphics cards, substantially boost the computing power of the RTX 3060/3090/4090, and significantly accelerate AI workloads such as image generation, model fine-tuning, and LLM inference.

KVCache Meetup Korea 2026, co-hosted by PyTorchKR and XCENA, featured talks from XCENA, SK hynix, FuriosaAI, and LMCache on CXL memory pooling, GPU/HBM roofline analysis, batch scheduling on RNGD, and process-isolated KV Cache serving.

A tweet sharing commentary on Ben wanting ordinary people to use AI as economic actors to prevent hedge funds from dominating the economy, and why that requires education.

A tweet criticizes a viral claim about 'Weight-Space Programming' where GPT-6 Astra supposedly edits Qwen3.5-0.8B weights to boost math performance, noting the results lack competent baselines and the Qwen result is mere bug fixing, while acknowledging that models reading/writing weights in toy problems is a real trend worth watching.

Fable 5.5 reportedly generated a complete animation in a single pass using only code. Chetaslua said that standard Opus 5.5 and Fable 5.1 produced noticeably worse results with the same prompt and provided a method for verification, but the report has not yet been officially confirmed.

The paper proposes Latent Sparse Attention (LSA), which hierarchically clusters Keys in latent space so that Queries only need to score against cluster centers. This achieves a 72.4x speedup in Prefill attention and a 42.5x speedup in Decode for a single attention layer of Qwen3.8-27B under 10M ultra-long context, reduces end-to-end first-token latency to about 1/9, and keeps accuracy essentially unchanged. The method includes two variants: distillable D-LSA and natively trained N-LSA, with code and distilled weights open-sourced.

Stanford researchers retested 50 widely cited findings about LLMs using 20 new models and found that 44% no longer held, 34% had significantly weakened, and only 16% still held, suggesting that LLM research findings may have a “shelf life” of only about 15 months.

The author shares a retrospective of using their own subagent-team skill to lead a Claude/Codex/etc. AI team through a major monorepo overhaul: cross-vendor model reviews offer the best ROI, rework volume depends on task risk rather than model tier, and they summarize a division of labor where low-tier models act as coordinators, strong models handle risky implementation, and top-tier models handle reviews.

Oracle, Microsoft, Meta, Amazon, and Google have committed to more than $1.3 trillion in future data center lease payments, roughly 70% of which were committed in the past four quarters, as the AI computing arms race turns into a high-stakes bet on long-term leases.

The author believes the most interesting paper of the past two years is persona vectors, and recommends reading it carefully five times. They also point out that research such as Weight-Space Programming and emergent misalignment is essentially just old wine in new bottles.

The AI short drama "Armed Entry into Song" achieved over 6 billion views with a small team of about ten people using a "one-person production" model, leveraging Seedance 2.5, AI dubbing, and JianYing for production, demonstrating a viable path toward industrialized AI content creation.