English 日本語

Tech News

Current developments in generative AI, software engineering, programming, and digital products, with links to original sources.

Claude Codeの週次制限は「表面上は増額、実質17%減」──HF侵害事件の最終報告書、grill-meスキル100万突破、96GBカード1枚で170Kコンテキスト

Qwen3.8-Flash-Next on a 64GB Mac and Four V100s — A Local AI Notebook, an Ontology for Playing 100 Characters, and a $0/Month LLM Router

Karpathy Reveals He "Hasn't Written a Single Line of Code Since December" — PLCBench for LLM Agents × Industrial Control, a 192GB Framework, and Real-World 350K-Token Operation

Karpathy Reveals He "Hasn't Written a Single Line of Code Since December" — PLCBench for LLM Agents × Industrial Control, a 192GB Framework, and Real-World 350K-Token Operationのカバー画像

Meta's Reported "Project OT" Plan to Replace Employees with AI — The State of Local Voice AI, and Redefining TDD for the AI Era

GLM-5.3-Flash NVFP4 Hits 97% on HumanEval — The Quantization "Apparent Bits" Problem, and a Warning on AI Compute Investment

GLM-5.3-Flash NVFP4 Hits 97% on HumanEval — The Quantization "Apparent Bits" Problem, and a Warning on AI Compute Investmentのカバー画像

Sony Music and Warner Sue Anthropic — The Future of the Cursor Migration, and a Pew Survey Finds Signs of AI Writing on Over a Third of English Pages

Sony Music and Warner Sue Anthropic — The Future of the Cursor Migration, and a Pew Survey Finds Signs of AI Writing on Over a Third of English Pagesのカバー画像

95% of China's Short-Form Dramas Are Now AI-Generated — Actors Hand Over Voice and Likeness to AI Before Layoffs, Google's WikiSkill Lets Agents Learn From Failure, and Major Open Models Hit by Critical Vulnerabilities

95% of China's Short-Form Dramas Are Now AI-Generated — Actors Hand Over Voice and Likeness to AI Before Layoffs, Google's WikiSkill Lets Agents Learn From Failure, and Major Open Models Hit by Critical Vulnerabilitiesのカバー画像

Major AI Players Warn of Cyber Crisis "Within Months" — Hy4 Retains Nearly Full Accuracy with Official 1-Bit Quantization, Debian Approves "Responsible Use" of GenAI

Major AI Players Warn of Cyber Crisis "Within Months" — Hy4 Retains Nearly Full Accuracy with Official 1-Bit Quantization, Debian Approves "Responsible Use" of GenAIのカバー画像

LAION Releases a 10-Million-Hour Open Video Dataset — Research Demonstrates Code Execution Inside Fortune 500 Firms via Agent Files, and Scientific American Questions Whether AI Tutor Alpha School Works

LAION Releases a 10-Million-Hour Open Video Dataset — Research Demonstrates Code Execution Inside Fortune 500 Firms via Agent Files, and Scientific American Questions Whether AI Tutor Alpha School Worksのカバー画像

GLM-5.3 Matches Fable 5 on Terminal Bench 4.0 — AI Loss-of-Control Incidents Surge, and Only 4 of 129 Skills Ever Got Called

GLM-5.3 Matches Fable 5 on Terminal Bench 4.0 — AI Loss-of-Control Incidents Surge, and Only 4 of 129 Skills Ever Got Calledのカバー画像