Naveed Ullah
Posts on U.S. fab/foundry ambitions, advanced lithography approaches, and the geopolitics of chipmaking.
AI Animation Breakthrough Enables Instant, Unlimited Experimentation
🤯 AI animation is entering a new era. Cascadeur is removing one of the biggest bottlenecks: token-based generation. ✅ AI Inbetweening ✅ Smart posing ✅ Physics-based motion tools All running locally. No credits. No queues. No waiting. The standout demo showed a simple blockout rig turned into a complex running animation with near-instant iteration. The real breakthrough isn't just better animation. It's unlimited experimentation.
Open‑Source AI Closing Gap, GLM Threatens Closed Models
🚨 Open-source AI is closing the gap fast. GLM 5.2 is reportedly outperforming Opus on multiple web development tasks. Even more interesting, Zhipu founder Jietang says they're targeting a FABLE-like model before Q1 2027. If GLM keeps improving and adds stronger vision capabilities,...
Markerless Motion Capture Turns Video Into Full Animation
No mocap suit. No markers. Just video. Epic Games' MetaHuman Animator Markerless Motion Capture plugin can turn regular video into full character performance data, including body movement, facial animation, and expressions. Combined with Epic’s new stylized sample project, creators can now learn...
Midjourney Launches Medical Division, Signaling AI's Healthcare Leap
🚨 Big move from @Midjourney. The company has announced Midjourney Medical, a new division focused on healthcare. AI is rapidly expanding beyond content creation and into industries that could have a real-world impact on millions of people. Could healthcare become AI's next major...
NVIDIA MotionBricks Delivers Real-Time Humanoid Motion
NVIDIA’s MotionBricks is a glimpse of where humanoid robotics is heading. 15,000 FPS motion generation, 2ms latency, and 350K motion skills from a single neural model. No animation graphs. No task-specific fine-tuning. Just real-time movement generated from high-level goals. This is now...
LuvBotics Aims to Make AI Companion Robots Mainstream
Another Chinese startup just joined the companion robotics race. LuvBotics is betting that AI-powered robots could become the next major consumer tech category.
Description Over Construction: Codex + Blender Shows Future
We're entering an era where describing what you want may be more important than knowing how to build it Codex + Blender is proof.
SurfLo Generates High‑Quality 3D Models From Few Photos
Turning a few photos into a clean 3D model used to be a frustrating process. SurfLo changes that. It can generate usable 3D meshes from just a handful of standard images while maintaining strong geometric consistency and predicting millions of oriented surface...
AI Advances From Images to Interactive 4D Worlds
Most AI models can generate images. The next generation is learning to understand entire worlds. Today, @MetaAI released three new research papers showing how large-scale generative models can reconstruct 3D scenes, reason about depth, and transform video into interactive 4D experiences. This is...
AI Evolves From Answering to Driving Full Research Workflows
Most people see NotebookLM as a note-taking tool. Google sees it as an AI researcher. The latest upgrade adds agentic capabilities, stronger reasoning, and new output formats, making complex multi-step research dramatically easier. We're quickly moving from AI that answers questions to AI...
Build Real Data Centers, Learn Internet Infrastructure
Most people use the internet every day but never see what powers it. 🎮 Data Center is a Steam game where you build a real data center from scratch. Buy racks, install servers, connect every cable manually, and manage customer traffic. It's one...
AI Image Editing, Not Generation, Drives Reve 2.0
The most important feature in @ReveAI's Reve 2.0 isn't image generation. It's image editing. Reve 2.0 automatically detects layers in generated images, letting you edit specific elements with simple prompts instead of regenerating everything. A powerful workflow: Generate → Edit → Turn the results...
Better AI Performance via Structured Workspace, Not Size
Everyone is racing to build bigger AI models. Harness-1 took a different approach: build a better system around the model. Instead of forcing an AI agent to remember every search, document, and verification step in its context window, Harness-1 gives it a...
Free Open-Source Editor Replaces CapCut Pro
I just canceled my CapCut subscription. Clypra is an open-source video editor that brings CapCut Pro features to your desktop for free. Multi-track timelines, frame-accurate trimming, local FFmpeg exports, zero watermarks, no accounts. Full control is back.
Miso One Delivers Human‑Like Emotion in 110 Ms
AI voices are getting scary good. MisoLabsAI just launched Miso One, an 8B parameter text-to-speech model that delivers human-like emotion with only 110ms latency. What makes it stand out? • Expressive, natural-sounding speech • Creator-quality narration • Open-source model weights • API access coming soon This is...