What’s New in Tech? AI Debates, Real‑Time Captions, Taichi 0.7.15, and More

Google Real‑Time Caption Feature Rolls Out in Chrome

On March 18, the real‑time caption feature became available in Chrome 89, using machine‑learning to automatically generate subtitles for audio and video, improving accessibility for users with hearing impairments.

AI Can Debate Humans – Project Debater Shows Promise

Researchers from IBM introduced Project Debater, an autonomous system that can engage in argumentative debate with humans, drawing on a 400‑million‑document knowledge base. Although humans won the competition, the system demonstrates potential applications in finance, law, public policy, and business decision‑making.

Taichi 0.7.15 Released

Taichi, a high‑performance graphics programming language embedded in Python, released version 0.7.15 with refactoring of TypedConstants, ASTBuilder, and new assertions to prevent index overflow in debug mode.

JD Technology “Smart Chain” Enables Cold‑Chain Traceability

JD Cloud launched the “JD Cold‑Chain Traceability Platform”, covering fresh produce, meat, and seafood, allowing end‑to‑end product tracking and future inter‑provincial code sharing.

Cloud‑Network Industry Publishes First Whitepaper

The 2021 Cloud‑Network Conference in Beijing released the inaugural “Cloud‑Network Industry Development Whitepaper”, outlining current status, technical characteristics, use cases, and future trends of China’s cloud‑network sector.

China Academy of Information and Communications Technology to Release Trusted Digital Assessment Framework

In March, the academy announced an upcoming “Trusted Digital Assessment Framework” to help enterprises evaluate digital transformation solutions.

CVPR 2021 – Single‑Image Depth Super‑Resolution via Cross‑Task Knowledge Transfer

The paper proposes a method that distills scene‑structure knowledge from color images during training, enabling high‑quality depth reconstruction from a single low‑resolution depth map at test time.

CVPR 2021 – AdCo: Adversarial Contrastive Learning

AdCo introduces an adversarial approach to generate learnable negative samples for contrastive learning, achieving state‑of‑the‑art results on ImageNet with far fewer negatives.