August 26, 2026
AI News
The AI news worth knowing this week.
Taiwan charges nine people for smuggling ‘high-end’ AI servers to China
Taiwanese prosecutors charged nine people for illegally exporting high-end AI servers, including Nvidia and Super Micro employees, to mainland China. The servers are believed to be B300 graphics processing units, previously banned from sale to China.
Read the original →Liquid AI open-sources Pipette, a benchmark for AI running on phones and laptops
Liquid AI, together with Artificial Analysis, open-sourced Pipette, a tool that measures how AI models actually perform on the device in your hand: model, quantization, runtime and hardware all tested together, across macOS, iOS, Windows and Android.
Read the original →World humanoid robot games show runners breaking records, bursting into flames
At the World Humanoid Robot Games in Beijing, sprinting robots beat Usain Bolt's 100m record while others fell, sparked and caught fire. The less flashy chore events, like folding laundry and cleaning rooms, matter more as a real test, and there the robots still lag far behind humans.
Read the original →Stability AI, maker of image generator Stable Diffusion, raises $76 million in fresh funding
Stability AI, creator of the open-source image generator Stable Diffusion, secured $76 million in new funding, bringing its total raised to $232 million. Investors include Universal, Sony and Warner Music, EA and AMD Ventures.
Read the original →OpenAI disrupts a new covert influence campaign from Russia
OpenAI banned a cluster of Russia-based ChatGPT accounts that were drafting pro-Kremlin posts for X, LinkedIn, Facebook, Substack and Telegram, pushed through a fake think tank posing as Israel-based. Engagement was low.
Read the original →Granite 4.2: How IBM built its first reasoning-focused open models
IBM's Granite 4.2 is its first reasoning-focused model family: 3B, 8B and 30B, Apache 2.0, with a 512K context window. The post walks through the training recipe, from pretraining on 15 trillion tokens to reinforcement learning where the larger models practice coding and terminal tasks in sandboxes.
Read the original →Dubbing & Open source
Open-source voice and dubbing tools worth a look.
stronghamjji/PersoDub
Video dubbing that runs entirely on your own computer, in your own voice. Nothing is uploaded anywhere, and it never stretches or squeezes the audio to force it to fit.
Read the original →
tashfeenahmed/freellmapi
A self-hosted gateway that pools the free tiers of 34 AI providers (Gemini, Groq, Cerebras, Mistral, OpenRouter, GitHub Models and more) behind one address, and switches to the next key when a quota runs out. One command points Claude Code at it. Two big releases this month.
Read the original →
SysAdminDoc/VideoSubtitleRemover
A desktop app that finds burned-in subtitles in a video and erases them, filling the gap so the background looks untouched. Two releases this week added a preview where you pick which text to remove. Windows only for now.
Read the original →
Akimiya-z/codex-guard
A quality gate for pull requests written by AI coding agents: it blocks leftover TODOs, leaked secrets, sloppy commits and failing CI before they reach the main branch.
Read the original →This week on X
What AI Twitter was talking about this week.
Google Gemini (@GeminiApp)
Back-to-school season is here and starting today, eligible college students can get a full year of Gemini on us: US students get 1 year of Google AI Pro at no cost, and in 140+ countries students get 1 year of Google AI Plus at no cost. (Personal account and a payment method required; claim by December 31.)
Read the original →Ornith (@ornith_)
Aloha! Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving strategies. It achieves state-of-the-art performance among open-source models of comparable size and delivers performance comparable to Claude Opus. (Benchmark numbers are self-reported by the lab, DeepReinforce.)
Read the original →NVIDIA AI (@NVIDIAAI)
Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark. NVIDIA AVO completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals. (Note: this is the public test set only, and AVO runs on top of Claude Opus 5, so the result comes from how the agent is built, not from a new NVIDIA model.)
Read the original →ClaudeDevs (@ClaudeDevs)
Computer use, the browser tool, the Skills API, and the Files API are now generally available on the Claude Platform. Automate work in applications that have no API with fewer round trips per task, and build Claude Managed Agents on versioned skills and reusable files.
Read the original →Thariq (@trq212)
a skill people at Anthropic have been using a lot recently: ELI5. Type "/eli5 <what you want explained>" and it will "explain like I'm someone who knows nothing about this topic, using a HTML artifact with big pictures and few words".
Read the original →Get the next issue by email
In any of 8 languages. Unsubscribe anytime.