中文 EN

Bubble's Brain - 2025-12-09

Daily Brief

AI News 2025-12-09

AI Daily Brief

Summary

OpenAI partnered with Deutsche Telekom to roll out ChatGPT Enterprise, while Google’s CEO said AI will reshape work across all levels.
ByteDance’s AI phone sold out; SpaceX’s valuation could surpass OpenAI; Huawei created a foundation model department.
Zhipu open-sourced an AI Agent (AutoGLM) and a multimodal model family (GLM-4.6V) to lower barriers and grow an open ecosystem.

Today’s AI News

  1. OpenAI partners with Deutsche Telekom to expand ChatGPT Enterprise adoption: OpenAI announced a partnership with major European telecom operator Deutsche Telekom, which will deploy ChatGPT Enterprise across the company to improve internal workflows and customer service efficiency. The move highlights how large telecoms are accelerating adoption of enterprise generative AI as foundational infrastructure for digital transformation in Europe.

  2. Google CEO Sundar Pichai: AI will fundamentally reshape the workplace: Sundar Pichai said AI will deeply reshape the workplace, affecting roles from entry-level to executives. He characterized AI as a disruptive technology that will eliminate some jobs while creating new opportunities, urging society to adapt. Pichai advised young people to follow personal interests rather than only “job safety,” and to learn how to incorporate AI into their work.

  3. ByteDance’s “Doubao phone” sells out and sparks resale hype: The first batch of 30,000 units of the “Doubao phone,” a collaboration between ByteDance and ZTE, reportedly sold out quickly and surged on second-hand markets with prices doubling. The ¥3,499 engineering-sample device deeply integrates the Doubao AI assistant to explore tighter coupling between AI and mobile OS flows, enabling actions like price comparisons, shopping, and work task handling. While still imperfect, the market response underscores the potential of AI phones; the company also addressed user privacy concerns.

  4. SpaceX valuation talks could reach $800B—potentially surpassing OpenAI: Reports say SpaceX is discussing a new round of share sales that could value the company at up to $800B, which would surpass OpenAI and make SpaceX the world’s most valuable unicorn. The deal is mainly a secondary share sale to provide liquidity for employees and early investors. SpaceX’s valuation reportedly doubled in five months, with 2025 revenue expected to hit $15.5B. The company also plans a potential IPO in the second half of next year, possibly packaging its launch business together with Starlink, which operates ~9,000 satellites.

  5. Huawei’s 2012 Lab establishes a “Foundation Model Department”: Huawei’s 2012 Laboratories reportedly set up a Foundation Model Department to strengthen its strategic investment in foundational AI and base model research. To support long-term development, Huawei is recruiting top AI researchers globally via its UAE team, prioritizing candidates with original breakthroughs. The move signals a more systematic expansion of Huawei’s push into general AI capabilities and core algorithms.

  6. Google’s experimental infinite canvas Mixboard adds “one-click PPT generation”: Google’s experimental infinite-canvas tool Mixboard received a major update by integrating its Nano Banana Pro image model, enabling one-click PPT generation. Users can create professional slide decks in minutes by prompting on the canvas, importing images or PDFs, or even sketching edits. The tool is now publicly available in the U.S. Google plans to ship team spaces and an API in 2025 to broaden use cases.

  7. Meta signs deals with major media outlets for real-time news Q&A in its AI chatbot: Meta announced agreements with CNN, Fox News, and six other major outlets to provide real-time news Q&A inside its AI chatbot. Using real-time news tech powered by Llama 4, Meta AI will generate news cards with clickable external links pointing back to publishers’ sites. The move marks Meta’s first large-scale copyright payments since it stopped paying for news, aiming to restore Llama 4’s reputation and increase engagement by embedding Meta AI deeper into its social apps.

  8. Ant Group launches Lingguang web assistant that can generate “flash apps” in 30 seconds: Ant Group released the web version of its general AI assistant Lingguang, highlighted by the ability to generate lightweight “flash apps” via natural language in about 30 seconds, with seamless data sync across web and mobile. The assistant targets workplace and study scenarios and supports multimodal outputs such as 3D models, audio, and maps—positioning itself as an AI tool with “low barrier, high efficiency,” and claiming full-code generation for multimodal content.

  9. Google’s Doppl adds an AI-generated “shoppable discovery feed”: Google added a “shoppable discovery feed” to its AI try-on experimental app Doppl, consisting entirely of AI-generated short videos—with synthetic models and scenes. Users can jump from AI outfit videos directly to brand websites to purchase, signaling Google’s first integration of an “AI short video + instant shopping” loop as it competes for social commerce attention against platforms like TikTok.

  10. Anthropic tests Claude Code integration with Slack: Anthropic announced a beta integration of its coding agent Claude Code with Slack, allowing software engineers to assign coding tasks, fix bugs, and generate pull requests directly inside Slack—reducing friction between discussion and execution. The product reportedly reached $1B+ in annualized revenue within six months, with customers including Netflix and Salesforce. Internally, employees reportedly spend about 60% of their work time with Claude Code, with productivity up by ~50%.

  11. Zhipu open-sources AutoGLM agent model to lower the barrier for AI phones: Zhipu open-sourced its core AI agent model AutoGLM, which can reliably complete complex mobile tasks like ordering food and booking flights. The release enables hardware vendors and developers to reproduce an AI assistant that can “understand the screen” and mimic human operations inside their own systems—substantially lowering the technical barrier for AI phones and pushing the ecosystem toward openness. AutoGLM supports 50+ high-frequency apps including WeChat and Taobao and offers both local and cloud deployment for privacy.

  12. Zhipu open-sources GLM-4.6V multimodal model family with native Function Calling: Zhipu open-sourced the GLM-4.6V multimodal family, including a 106B base model and a 9B lightweight model. It natively integrates Function Calling into a vision model, connecting “visual perception” to “executable actions.” The lightweight GLM-4.6V-Flash is free and commercially usable; the base model’s API pricing is cut by ~50% vs the previous generation and is about a quarter of GPT-4V’s cost. The goal is to accelerate adoption of a low-cost yet powerful visual agent ecosystem.

Comments