✔️ OpenAI is developing its own internal platform to replace GitHub.
The company has begun working on a proprietary system for code hosting, aiming to reduce its dependence on Microsoft infrastructure. This move was prompted by recent outages on GitHub, which caused OpenAI engineers to lose access to their repositories and halted workflows.
Currently, the project is in the early stages, with a full launch planned within the coming months. At this point, it is only a corporate solution — the company has no plans to offer access to third-party developers.
Sources: theinformation.com
✔️ Physical Intelligence has developed a two-tier memory architecture for robots.
This startup announced Multi-scale Embodied Memory — a memory system designed specifically for VLA-model robots. The technology addresses the issue of losing context during complex multi-step tasks.
The system incorporates two mechanisms. Short-term memory operates with an optimized video encoder — providing spatial awareness and enabling models to quickly adapt to errors during execution. Long-term memory uses text-based abstract notes — the model creates linguistic notes about its actions, recording task status and the location of hidden objects.
Integrating MEM into the base model π0.6 allowed robots to successfully handle tasks lasting up to 15 minutes. This approach lays the foundation for autonomous systems capable of planning complex household scenarios hours or even days in advance.
More details: pi.website
✔️ Autodesk has introduced a new AI-powered generation tool for creating 3D models.
The company added Wonder 3D, a generative tool, to the Flow Studio platform. It significantly speeds up prototyping: now users can create complete 3D characters and objects within minutes based on text descriptions or reference images.
Additionally, Wonder 3D not only generates basic geometry but also offers utilities for asset refinement: applying textures, reconstructing mesh topology, and editing individual elements without needing to regenerate the model.
Created objects can be easily exported to popular 3D editors, game engines, or 3D printing software. These tools are available across all Flow Studio plans, including the free basic tier. Each request costs a fixed 20 credits.
More information: autodesk.com
✔️ NotebookLM now features Cinematic Video Overviews.
Google expanded its NotebookLM product with a new tool — the Cinematic Video Overviews generator. It automatically creates high-quality animations, replacing traditional voice-over presentations.
The new feature is based on the integration of Gemini 3, Nano Banana Pro, and Veo 3. Gemini 3 acts as a virtual creative director — determining the style, structure, and storyline of the upcoming video without user input.
Initially, this feature supports only English and is available exclusively to Google AI Ultra users.
More details: NotebookLM
✔️ Alibaba’s key Qwen team departs after the release of Qwen 3.5 series models.
Just hours after announcing the new compact models, the Qwen 3.5 series lost several leading members: CTO Junyan Lin — who oversaw all Qwen generations; researcher Binyuan Hui; and developers Kaisin Li and Yu Boyen.
The official reason cited is internal reorganization within Tongyi Lab: moving from product team consolidation to creating specialized working groups. However, reports suggest that the departure of Qwen creators is related to low daily user activity metrics for these products.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby