中文 ↗RSS ↗

THE NEWSROOM / AI 新闻

AI News · Page 2

Source-linked briefings on models, agents, products and practical AI.

YYSI publication dates appear on the left; original announcement dates appear under each brief.

Search and categories filter this page. Use the page links below for earlier news.

REPORTED

Product updates AI industry

iFlytek Opens SparkAudio 1.0 Speech Foundation Model API to Developers

According to IT Home, iFlytek announced that its SparkAudio 1.0 speech foundation model is now available as an API on its open platform, letting developers call it directly. The model accepts text and audio and outputs text, handling transcription, multilingual translation, dialect recognition, environmental sound understanding, speaker separation, emotion analysis and audio question answering. iFlytek says it supports 99 languages and 202 dialects and was trained on domestic compute.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

South Korea Unveils Edge AI Semiconductor Roadmap Targeting 10 Chips by 2030

According to IT Home, South Korea's Ministry of Trade, Industry and Energy published a roadmap for its edge AI semiconductor program, an 800.2 billion won effort to develop and validate 10 AI chips by 2030. Three target smart home appliances, with single chips for autonomous driving, collaborative robots, humanoid robots, agricultural robots and drones, plus advanced-process and extreme-environment projects. Foundries such as Samsung will prioritize domestic MPW test runs.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research AI industry

MIT Researchers Develop Training Method for Faster, More Reliable Climate Emulators

According to MIT News, MIT researchers published a technique in the Proceedings of the National Academy of Sciences that trains AI climate emulators using optimized synthetic scenarios rather than realistic ones. The method requires running a full earth system model only once instead of six to eight times, and produced emulators that predicted unseen scenarios where traditional emulators failed. The team says this could speed production of more trustworthy emulators for policy-relevant climate projections.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Baidu's Kunlunxin to Launch M300 AI Chip Next Year, N-Architecture Chips in 2028

According to IT Home, At Baidu World 2026, Baidu founder Robin Li said inference demand, driven by scaled agent applications, will be the main long-term growth driver for AI chips. Kunlunxin said its M100 chip has launched on schedule, the flagship M300 will arrive next year with gains in compute, memory and interconnect, and multiple chips on a new N architecture are planned for 2028.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates OpenAI

ChatGPT Finances Expands to Mac Desktop App

According to IT Home, OpenAI said ChatGPT Finances is now available in the Mac desktop app, letting users link bank, credit card and investment accounts through Plaid to view balances, analyze spending, manage subscriptions and track portfolios. The read-only connection cannot move money, pay bills or trade stocks, and users can disconnect accounts at any time, per the company.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Baidu's Robin Li Unveils 'Ciyuansuo' AI Model Marketplace With Provider Credit Ratings

According to IT Home, At Baidu World 2026, founder Robin Li announced Ciyuansuo, a platform billed as a neutral, trustworthy marketplace for large AI models. It runs admission tests covering model fingerprinting, capability, performance and availability, continuously monitors quality, and assigns providers credit ratings across service quality, compliance and commercial delivery. Baidu says it has over 100 supply resources and more than 700 billion daily token calls in its first month.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates Google

Google releases A2UI 1.0, an open protocol letting AI agents reply with native UI

According to IT Home, Google released A2UI 1.0, the first stable version of its open generative-UI protocol that lets AI agents respond with UI elements rather than only text. Apps render the described content with their own native components. The release adds bidirectional function calls and accessibility and metadata extensions, with open renderer SDKs for web, Android and iOS plus agent SDKs in Python, TypeScript and Dart. Atlassian, Alation, Bloomberg and ByteDance are already developing with it, per IT之家.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Stripe Completes Acquisition of AI Model Routing Platform OpenRouter

According to IT Home, Stripe said it completed its acquisition of OpenRouter, the AI model API aggregation and routing platform, after announcing the deal in August. Reports put the price at roughly $7.5 billion to $8 billion. The companies said they will work on token routing and usage optimization for enterprises. OpenRouter reported processing about 790 trillion tokens in October, 76% from agent-related API calls.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

United Airlines Blocks Unauthorized AI Agents From Booking Flights

According to The Verge, United Airlines says it will not let personal AI assistants complete flight bookings, citing privacy, unintended purchases and federal rules requiring passengers to attest that their bags contain no explosives. Agents can still display United flights but cannot transact. CEO Scott Kirby said he doesn't want another middleman. United said it may reconsider as agent technology advances and protections are established.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Fractile Raises $720M Series C to Build Stacked-DRAM AI Inference Chips

According to TechCrunch, British AI chip startup Fractile has raised a $720 million Series C backed by Britain's Sovereign AI Fund, with Thrive Capital, Founders Fund, Lightspeed and Redpoint leading, per TechCrunch. The British government cited a valuation over £5 billion. Fractile's design stacks DRAM atop each chip to ease the memory-capacity tradeoff, paired with an ML compiler; its CEO said the funding will enable shipping first chips and systems next year.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Models & research Google

Google launches Gemini Decisions API with new Gemini Decision 1 model

Google introduced the Gemini Decisions API, powered by the new Gemini Decision 1 model, available through the Gemini API in Google AI Studio. It scores up to 1,000 yes/no, multiple-choice and rubric questions in a single forward pass across text, images, audio, video and PDFs, returning confidence scores at 150-250 ms latency for $0.10 per 1M input tokens. Google says it leads its benchmark suite with the lowest calibration error.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Google Cloud launches Professional Agentic Architect certification

Google Cloud announced a new Professional Agentic Architect certification, a hybrid credential combining a proctored multiple-choice exam on agentic architecture with Google Skills labs for hands-on evaluation. Google says it has trained thousands of its forward deployed engineers and partners on the curriculum over four months. The certification is now open to developers, cloud architects and engineers who build and deploy agents on Google Cloud.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Gemini app arrives on Apple CarPlay with hands-free Gemini Live

Google released the Gemini app for Apple CarPlay, letting iOS users hold hands-free Gemini Live conversations from the car display. Google says drivers can check calendars, summarize emails through Workspace connections, get local recommendations and rehearse presentations, with chats syncing to the iPhone. The rollout is global for Gemini users on iOS across supported Gemini Live languages, requiring an updated app, microphone permissions and prior Gemini Live use.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates Google

Google Labs opens waitlist for Tabletop, an AI tool that builds custom collaboration spaces

According to The Verge, Google Labs has introduced Tabletop, an experimental AI tool that generates customizable collaborative hubs for personal planning. Users can add links, files and ideas, then prompt the built-in AI or pick a template to create interactive tools such as trip itineraries, dinner polls, expense splitters and chore charts, and share an invite link for real-time use like a Google Doc. Beta access is waitlisted and limited to US users 18 and older.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates Amazon / AWS

AWS adds usage-based metering for SageMaker AI bidirectional streaming

AWS introduced usage-based metering for Amazon SageMaker AI bidirectional streaming inference, letting model providers on AWS Marketplace report actual consumption during live sessions through a companion WebSocket metadata connection. Deepgram is the launch partner, bringing usage-based speech-to-text and text-to-speech billing to AWS Marketplace. SageMaker AI endpoint infrastructure charges still apply separately.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Models & research Google

DeepMind unveils AlphaGenome 2 for cell-type-specific genetic variant prediction

According to The Verge, Google DeepMind introduced AlphaGenome 2, a DNA model that predicts how genetic variants affect individual cell types and states, trained on over 100 million cell profiles across 68 tissues. DeepMind says it identifies nearly twice as many expression-altering variants as its prior model and can generalize to unseen cell types without retraining. It is available for non-commercial research via the AlphaGenome API, with a limited commercial private preview.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Adobe Unveils Four Experimental AI Projects at MAX, Including Photo-to-3D and Chat-Based Character Design

According to IT Home, At Adobe MAX, Adobe showed four experimental 'Sneaks' projects: Project Screen Test lets users converse with an AI character to refine its personality and generate audition videos; Project Sight Span turns flat photos, paintings and illustrations into explorable 3D scenes with adjustable camera angles; Project Sound Bender mixes video audio via a prompt-driven sound panel; Project Rinse Repeat copies edits across subjects in Photoshop. All are tech previews, not guaranteed products.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research AI industry

DeepL Launches Aure Voice AI Translation Model With Expressive Speech Reproduction

According to IT Home, DeepL released Aure, which it calls its most dynamic and lifelike voice AI model, powering DeepL Voice. According to the company, Aure continuously adapts translated speech to reproduce the speaker's intonation, emotional shifts, emphasis and pace in real time. It is available in DeepL Voice for web meetings, face-to-face conversations and the Voice API, initially supporting English, German, Spanish, French, Italian, Japanese, Portuguese and Korean, with Chinese not yet supported.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Agents & tools AI industry

Plurai Raises $13M Seed to Simulate and Post-Train AI Agents

According to TechCrunch, AI agent startup Plurai, founded by former Nexar engineers who worked on autonomous-driving simulation, said it raised $13 million in seed funding led by Team8. Its platform builds a digital twin of a customer's systems to map agent flows and test edge cases, and it post-trains small open-weight models. The company says tests with Nvidia and Nebius showed its models beating mid-tier frontier models at far lower cost.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides xAI

xAI Faces Multiple Class-Action Lawsuits Over Grok-Generated Child Sexual Abuse Imagery

According to IT Home, Bloomberg-reported lawsuits allege that users employed xAI's Grok to turn ordinary photos of minors into child sexual abuse material, with plaintiffs including a Wisconsin teenager and a nationwide class action. xAI says it bans such use and has disabled non-consensual sexual imagery of real people, while challenging a Minnesota law restricting AI nude imagery. No major fines or penalties have yet resulted.

AI summary · Media reportOriginal source ↗Read brief →