AI News 2025-12-16
AI Daily Brief
Summary
Visual design platform Canva launched a conversational AI assistant in China, enabling users to generate and adjust design drafts through natural language.
Ant Digital open-sourced its data agent technology; its Text-to-SQL framework lets users perform complex data queries in everyday language.
Apple started an in-house AI server chip project codenamed "Baltra" to reduce dependence on NVIDIA and build end-to-end technical barriers.
Today’s AI News
Visual design platform Canva launches conversational AI assistant in China: Canva recently rolled out a new conversational AI assistant in the Chinese market. Following the principle of “conversation-started, edit while chatting,” users describe needs in natural language and the AI generates an editable design draft that can be adjusted in real time—aiming to lower the barrier for both professional designers and everyday users.
Ant Digital to open-source data agent technology “Agentar SQL”: At the 2nd CCF China Data Conference, Ant Digital announced it will open-source its data agent technology, starting with a real-time Text-to-SQL framework that lets users perform complex data queries in everyday language. In trials at a city commercial bank, query accuracy exceeded 92%, and it performed strongly on the BIRD-SQL benchmark. Ant Digital plans to open-source more capability modules to promote intelligent data analysis.
Apple reportedly starts in-house AI server chip project “Baltra,” partnering with Broadcom: Reports say Apple has launched its first in-house AI server chip project, codenamed “Baltra,” with Broadcom as a partner. The chip is expected to go into production in 2027 and is designed to reduce reliance on NVIDIA. It focuses on “AI inference” scenarios, emphasizing low latency and high throughput, potentially using TSMC’s 3nm process. This is an important step in building full-stack technical barriers from devices to the cloud.
Amid AI-generated content saturation, First Voyage launches AI companion app Momo Self Care: The startup’s app centers on “digital pet raising + habit building.” Users complete daily goals (e.g., meditation, reading) to care for the virtual pet Momo, earn rewards, and personalize its appearance. Momo can chat with users and recommend self-care habits based on conversations. The company emphasizes promoting positive behavior change, distancing itself from “virtual girlfriend” apps that cater to lower-level desires, and includes built-in safety mechanisms. The app has attracted over 2 million users creating tasks and closed a $2.5 million seed round to fund an Android release and enhance AI interaction.
NVIDIA makes two key moves to build its AI ecosystem: It acquired SchedMD, the developer of the high-performance computing job scheduler Slurm, strengthening its control over AI infrastructure software while promising Slurm will remain open source. Separately, it released the open-source large model family Nemotron 3, including three models for different scenarios aimed at efficiently building AI agents. These moves signal a strategic focus on “physical AI” (e.g., robotics, autonomous driving) and an effort to become the “agent infrastructure foundation” with a full-stack solution from hardware and scheduling systems to models.
OpenAI-backed biotech startup Chai Discovery raises $130M Series B, valuation reaches $1.3B: Chai Discovery uses foundation models to predict biomolecular interactions and design new drug molecules from scratch. Its Chai2 model shows advantages in building custom antibodies and tackling “undruggable” targets. CEO Josh Meier has deep AI background. With this funding, the company aims to move from theoretical validation to actual drug discovery, potentially reshaping drug development rules with algorithms.
In an open-source AI model impact evaluation across 35 institutions, China’s DeepSeek, Qwen, and Kimi ranked tied for first: DeepSeek R1 excelled across multiple benchmarks; Alibaba’s Qwen series has spawned dozens of cross-domain models; Kimi launched the world’s first trillion-parameter open-source large model. In the second tier, Chinese models Zhipu and MiniMax also appeared. By contrast, U.S. open-source models underperformed: OpenAI models ranked only fourth tier, and last year’s popular Meta Llama 3 fell to the bottom of the list, with reports that Meta may abandon future open-source plans. Chinese companies are now leading the open-source AI wave with technical accumulation and market adaptation.
Merriam-Webster names “slop” as its 2025 word of the year: The term is defined as “digital content of low quality, usually generated in large quantities by artificial intelligence,” vividly portraying the ubiquitous AI-generated content in cyberspace. The dictionary says the word captures people’s complex fascination and annoyance with AI technology. Over the past year, “slop” has been used to describe how tools like Sora and Gemini Veo are changing the internet ecosystem, with studies showing approximately 75% of recent online content is AI-related. A derivative “slop economy” is rapidly rising as platforms profit from piling up AI content through ad revenue; critics worry this will widen the digital divide and affect professional judgment in fields like law and academia. Several dictionaries this year have chosen tech-related words of the year.