New AI Thought Machine Unveiledđź§ đź’ˇ

Sakana’s CTM thinks like a brain, Meta’s AssetGen 2.0 builds rich 3D worlds, and Microsoft Pages brings smart, collaborative writing to life.

The Future of AI Is Taking Shape — From Thought to 3D Worlds to Collaborative Words . AI is evolving beyond static prompts and simple outputs—into systems that think, build, and write more like humans. Sakana AI’s Continuous Thought Machine introduces brain-inspired reasoning with time-aware neurons and step-by-step problem solving. Meta’s AssetGen 2.0 revolutionizes 3D content creation, generating detailed meshes and high-quality textures from simple prompts, bringing the metaverse closer to reality. Meanwhile, Microsoft’s Pages empowers writers with collaborative editing tools, blending human creativity with AI assistance inside the familiar Microsoft 365 environment. Together, these breakthroughs hint at a future where AI doesn’t just respond—but reasons, builds, and collaborates.

Continuous Thought Machine: Brain-Inspired AI

The Continuous Thought Machine (CTM) is a new AI model developed by Sakana AI that takes inspiration from how biological brains work. Unlike traditional AI, which treats neurons as simple units, the CTM allows each neuron to process a history of signals over time, enabling more detailed and flexible thinking. It also uses neural synchronization, meaning it encodes information in the timing of neural activity, similar to how real brains operate. This approach helps the CTM to think through problems step-by-step, showing human-like reasoning and planning. It works on both static and sequential data by operating in continuous time. Tests have shown the CTM can solve complex tasks like navigating mazes and classifying images while mimicking human attention and memory. By combining ideas from biology and AI, the CTM offers a new way to build smarter, more adaptable artificial intelligence.

Meta AssetGen 2.0: Next-Gen 3D AI

Meta has introduced AssetGen 2.0, a significant advancement in 3D generative AI designed to create high-quality 3D assets from text and image prompts. This new foundation model consists of two complementary AI systems: one generates detailed 3D meshes using a cutting-edge single-stage 3D diffusion model, and the other, called TextureGen, produces high-resolution, production-ready textures with improved view consistency and texture in-painting capabilities. Compared to its predecessor, AssetGen 1.0, AssetGen 2.0 delivers dramatically improved geometric detail and fidelity by being directly trained on a large corpus of 3D assets. Currently used internally by Meta for building 3D worlds, AssetGen 2.0 will be rolled out later this year to creators on Meta’s Horizon platform, enabling artists, designers, and developers to create complex 3D objects and environments more easily and accessibly-akin to 2D content creation. Furthermore, Meta plans to extend this technology to auto-regressively generate entire 3D scenes by sequentially creating individual objects and elements from simple prompts, marking a transformative step in 3D content creation and democratization for virtual reality and metaverse development.

Microsoft Pages: AI-Powered Collaborative Writing Too

Microsoft Copilot offers protected AI experience for UO | OregonNews

Microsoft has launched Pages, a new feature in its Copilot suite designed for collaborative writing and text refinement, offering an experience similar to ChatGPT’s Canvas but without coding capabilities. With Pages, users can directly edit AI-generated responses by highlighting text and using the “Ask” button to prompt modifications such as tweaking tone, expanding sections, or rewording ideas. This tool aims to streamline content iteration and make it easier for users to polish their writing collaboratively, but it currently does not support code editing or generation. Pages is positioned as a user-friendly editing assistant, allowing for real-time collaboration and seamless integration with other Microsoft 365 tools, further enhancing productivity and document quality for writers and teams.

Hand Picked Video

In this video, we'll look at o1-preview, o1 Beta, o1 preview API.

Top AI Products from this week

  • Probo - Probo is an open-source platform for SOC 2, ISO 27001, HIPAA, and more. If you don't want to lose time doing it, we provide a white-glove service to run everything for you and let you focus on your business. Open-source, transparent, and stress-free.

  • Syft AI - Unlike other news apps, you tell us the exact topics you care about, and we bring you time-saving, unbiased takeaways from the world's most trusted sources.

  • AI Meeting Notes by Notion - Focus on the conversation while AI captures every word. Meeting notes are always right where you work, so you can skip the recap and get to work faster. Works on any device and connects with your calendar.

  • Asendia AI - The first AI career agent that helps jobseekers practice with AI mock interviews, get personalized feedback, track job applications and land dream jobs. With the power of conversational AI, Asendia AI helps job seekers showcase skills and connect globally .

  • AI Avatar Videos - Create hyperrealistic avatars with natural gestures and perfect lipsync. Powered by ZERO SHOT AI LIPSYNC, this means that you don't have to submit any training videos if you want to use to create your AI Clone. Make your AI Avatar videos in minutes.

  • RADiCAL - RADiCAL is a real-time, browser-based platform for AI-powered motion capture and 3D visualization. Capture motion with a phone, animate 3D characters instantly, and collaborate in the cloud. No suits, markers, or software downloads needed. Get started for free

This week in AI

  • Single-Image to 3D Avatar - SVAD generates high-fidelity, animatable 3D avatars from just one image by using video diffusion and data augmentation, ensuring identity consistency and real-time rendering.

  • Self-Evolving AI Reasoning - Absolute Zero trains AI to propose and solve its own tasks, learning through self-play-no external data needed. AZR achieves SOTA math and coding results, surpassing data-trained models.

  • Meta AI Lab Gets New Chief - Robert Fergus, ex-Google DeepMind director, now leads Meta’s FAIR lab after Joelle Pineau’s exit, aiming to revitalize core AI research amid staff departures.

  • Gemini 2.5 Pro: Video AI Leap - Gemini 2.5 Pro leads in video understanding, surpassing GPT-4.1 on benchmarks. It powers interactive apps, animation from video, and precise moment retrieval at scale

  • AI Friends Zuckerberg’s Bold Future - Zuckerberg envisions most friends as AI-companions, therapists, and agents-embedded in daily life to fill social gaps and boost connection, not replace real friendships.