What’s New This Week? 3D File Specs, AI Benchmarks, Cloud PC & More

New 3D Data File Specification 1.0 Released

Jointly developed by JD, Huawei, Alibaba, Tencent, NetEase, Cocos, and other partners, the "New Generation 3D Data File Unified Technical Specification 1.0" defines a unified 3D file format covering basic structure, efficient editing, storage, parsing, fast data loading, engine‑agnostic interfaces, full scene representation (including nodes, transforms, networks, materials, cameras, animations), and extensible custom properties.

Flash Says Goodbye – Adobe Player Discontinued Jan 1

Adobe will cease support and updates for Flash Player on January 1, urging users to uninstall by January 12. Flash, first released in 1996, enabled creators to share animations and games over dial‑up internet.

deepin OS 20.1 Released

deepin 20.1 (1010) upgrades to Kernel 5.8 and Debian 10.6, improving system stability and compatibility. The release also updates the deepin suite and optimizes startup time, installation time, resource usage, and responsiveness.

China Telecom Launches First Quantum‑Secure Call Product “Quantum Secret Talk”

The service uses quantum random numbers and quantum key distribution to generate authentication and call keys, providing end‑to‑end encryption for voice calls. China now operates the world’s largest quantum communication network, moving toward commercial applications in government, finance, healthcare, defense, and more.

AI Needs New Benchmarks – Turing Test Is Outdated

Experts argue that the classic Turing Test no longer reflects modern AI capabilities. With chatbots now able to fool over 30% of judges, the old threshold is considered arbitrary, prompting a call for new evaluation metrics.

Windows 10 Cloud PC Upgrade Confirmed

Microsoft’s upcoming Windows 10 Cloud PC, built on Azure and Windows Virtual Desktop, will offer three configurations (general, heavy, premium) and integrate Graph v1.0 APIs for real‑time collaboration, signaling a shift toward cloud‑based desktop experiences.

AAAI 2021: YOLObile Achieves 19 FPS Real‑Time Detection on Mobile

A model‑compression and compilation framework compresses YOLOv4 by 14×, enabling 19 FPS inference on smartphones while maintaining 49 mAP on the COCO dataset—seven times faster than full YOLOv3 without sacrificing accuracy.

RealFormer Improves Transformer Attention by Shifting Residuals

RealFormer moves residual connections into the attention matrix, combining the training friendliness of Pre‑LayerNorm with the performance advantages of Post‑LayerNorm, effectively closing the gap between the two approaches.