中文 EN

Bubble's Brain - 2025-12-06

Daily Brief

AI News 2025-12-06

AI Daily Brief

Summary

The European Commission formally opened an antitrust probe into Meta for restricting third-party AI access to WhatsApp.
Meanwhile, OpenAI and Microsoft released new developer and speech models, and new data highlighted a surge in global unicorn valuations led by the U.S. and China.

Today’s AI News

  1. EU opens antitrust investigation into Meta over WhatsApp AI access restrictions: The European Commission formally launched an antitrust investigation into Meta after WhatsApp Business API policy changes reportedly allowed only Meta AI while blocking third-party general-purpose chatbots like ChatGPT. The EU alleges Meta is abusing market dominance and creating unfair competition. If found in violation, Meta could face fines up to 10% of global annual revenue—roughly $16.45B based on 2024 revenue. Meta argues the restriction prevents system overload and says competition in AI remains robust. OpenAI and Microsoft have announced they will remove WhatsApp entry points, while the investigation outcome could force WhatsApp to open its API and set a regulatory precedent for AI policies in other super-app platforms.

  2. Kling AI launches “Subject Library” to keep characters and objects consistent in AI videos: Kling AI introduced a Subject Library feature to address inconsistency in AI video generation—where characters, objects, and scenes can drift across shots. Users create a subject by uploading a small set of reference images; the system auto-completes multi-view representations and generates descriptions, giving the model a form of “memory.” During creation, users can call subjects via “@subject-name” to keep appearance consistent across any camera angle, and multiple subjects can interact stably in one scene. Kling also provides an optimized official subject library for quick reuse.

  3. 2025 Global Unicorn 500 report: total valuation up 30.7%, dominated by the U.S. and China: The 2025 Global Unicorn 500 report released in Qingdao said global unicorns reached a total valuation of ¥39.14 trillion, up 30.71% year-over-year. The U.S. and China dominate in both count and valuation—together accounting for 74.8% of companies and 86.8% of total valuation. The U.S. ranks first with 224 companies valued at ¥18.97 trillion, concentrated in AI and fintech; China follows with 150 companies valued at ¥12.83 trillion, showing strength in advanced manufacturing and auto tech. The report also notes increasing concentration in top cities like San Francisco and Beijing.

  4. Amap launches “AI Parking Radar” to predict street parking availability in minutes: Amap released AI Parking Radar, using spatial intelligence sensing and AI vision analysis to predict and dynamically display street parking availability at minute-level granularity. Drivers can view parking status along a route while navigating, easing parking pain. The feature first rolls out across tens of thousands of street parking spots in Beijing, signaling navigation apps expanding toward end-to-end mobility services.

  5. OpenAI ships GPT-5.1-Codex-Max for developers with agentic coding and ultra-long context: OpenAI released a stronger, same-price programming model GPT-5.1-Codex-Max for developers. The model offers agentic coding and ultra-long context processing—working with 1M+ tokens and sustaining tasks for 24+ hours—and reports better benchmark performance than previous versions. It’s optimized for Windows development workflows and is integrated into tools such as GitHub Copilot, widely seen as a strong response to competitor Anthropic in the premium developer tooling space.

  6. Volcano Engine releases Doubao Speech Recognition 2.0 with multimodal understanding: Volcano Engine launched Doubao Speech Recognition Model 2.0, improving inference and accuracy in complex scenarios and adding multimodal understanding—combining text and image context to reduce recognition errors. It also supports 13 overseas languages including Japanese and Korean. The model is available via the Volcano Ark Experience Center and API.

  7. Microsoft open-sources VibeVoice-Realtime-0.5B: 300ms latency, 90-minute speech generation: Microsoft open-sourced a real-time speech model VibeVoice-Realtime-0.5B, which can start speaking within about 300 ms. Despite only 0.5B parameters, it can generate up to 90 minutes of fluent audio and natively supports up to four speakers, each with distinct voice and emotion. It can automatically express emotions like apology, excitement, and anger based on text. VibeVoice supports Chinese and English (English near commercial quality; Chinese still improving) and is fully open-sourced on HuggingFace and GitHub with commercial use allowed.

  8. Vidu releases Q2 update: image generation + editing in a one-stop workflow: Video generation platform Vidu (by Shengshu Tech) released Q2, integrating reference-based image generation, text-to-image, and image editing. The update improves fine control of position, motion, and composition, supports 4K output, and adds editing features like local inpainting and material replacement—forming an end-to-end workflow from images to video. Vidu also launched a limited-time promo through Dec 31, offering flagship members unlimited image generation.

Comments