Alibaba's International Digital Trade Group has launched Ovis2.5, a new multimodal language model offering 9B and 2B parameter versions. Designed for economic visual reasoning, it sets new benchmarks in performance and efficiency, with applications in document analysis, OCR, and complex diagram interpretation.
Following the controversial launch of GPT-5, OpenAI restored access to GPT-4o after recognizing users' strong emotional attachment to the older model. ChatGPT head Nick Turley admitted gaps in understanding user needs, while CEO Sam Altman highlighted AI's psychological support role. The incident reveals challenges in balancing innovation with user expectations.
Autonomous vehicle leader WeRide announces a multi-million dollar investment from Grab to deploy thousands of L4 autonomous taxis across Southeast Asia. The partnership combines WeRide's self-driving technology with Grab's platform network, marking a major step toward commercial driverless mobility services in the region while addressing driver transition programs.
Oracle and Google Cloud have announced a deepened partnership to integrate Google's advanced Gemini AI model into Oracle Cloud Infrastructure (OCI). The collaboration aims to enhance AI-driven solutions across industries, including multimodal understanding, coding automation, and enterprise workflows. The integration will also leverage Vertex AI for generative AI services, offering customers more flexibility in deploying AI solutions.
The Zed editor team's analysis argues that while AI assists coding, it lacks the mental modeling capabilities of human engineers. Their blog post sparked debate about LLMs' limitations in software architecture and problem-solving.
Multiverse Computing, a Spanish AI startup, has launched two ultra-small AI models—'Fly Brain' (94M parameters) and 'Chicken Brain' (3.2B parameters)—using quantum-inspired compression. These models enable offline intelligent conversations on IoT devices and outperform larger counterparts in benchmarks. The company is in talks with Apple, Samsung, and HP for integration.
Kuaishou's Qilin 2.1 video generation model introduces a new keyframe function, improving video quality and smoothness. The upgrade allows precise control over video transitions, making it ideal for professional scenarios like commercials and AI films. The model also reduces generation time and costs significantly.
Alibaba Group's Intelligent Information Business Group has announced plans to recruit nearly 1,000 AI professionals, focusing on large language models, multimodal AI, and hardware development. The hiring initiative spans multiple Chinese tech hubs and includes both experienced researchers and fresh graduates.
OpenAI has reinstated access to GPT-4o after users expressed dissatisfaction with the newer GPT-5 model. Nick Turley, head of ChatGPT, noted the unexpected popularity of GPT-4o and emphasized the company's commitment to user feedback. The decision highlights challenges in balancing technological advancements with user experience.
MuleRun, an emerging AI platform, provides users with dedicated virtual machine agents capable of automating tasks from gaming to 3D modeling. Its unique ecosystem combines VM technology with community-shared automation scripts, offering unprecedented flexibility in AI applications.
Meta has open-sourced DINOv3, a self-supervised AI vision model that eliminates the need for manual annotations. The model excels in high-resolution feature extraction and supports diverse applications, from environmental monitoring to healthcare. Its release under a commercial-friendly license is set to democratize access to cutting-edge vision technology.
SUITONG Technology's humanoid robot H1 secured the first-ever gold medal in a 1500-meter race at a global humanoid robot competition. The event, held at Beijing's National Speed Skating Oval, featured over 500 robots from 16 countries. The H1, previously seen on China's Spring Festival Gala, demonstrated upgraded speed and endurance capabilities through specialized algorithm adjustments.
The UK government has launched a £4 million AI-driven crime mapping initiative to predict and prevent offenses like knife crime. The system will integrate police and social data to create real-time risk assessments, though concerns remain about privacy and potential bias.
The Doubao app has launched a new feature allowing users to generate animated emoji effects from static photos with one click. The tool offers various templates and preserves facial details, enabling creative social media content and expressive messaging.
IDC predicts China's generative AI hardware investments will skyrocket from 6.5B yuan in 2023 to 33B yuan by 2028, driven by explosive growth in token consumption and high-end network infrastructure demands.
Tencent Yuanbao has introduced a new feature allowing users to purchase physical books directly from JD.com. When book titles appear in Yuanbao's responses, underlined links will redirect users to JD.com's purchase page. This enhancement complements existing integrations with platforms like WeChat Reading and Qidian Reading, streamlining the journey from content discovery to purchase.
A new AI evaluation benchmark, FormulaOne, revealed that leading models like GPT-5 scored zero on advanced reasoning tasks. Developed by AAI, the benchmark tests dynamic programming problems at doctoral-level difficulty, exposing limitations in current AI capabilities.
OpenAI is exploring the introduction of ads to ChatGPT, as revealed by executive Nick Turley. Despite surpassing 700 million users, the company remains unprofitable. While cautious about disrupting user experience, OpenAI is weighing commercial strategies, including subscriptions and a new 'ChatGPT commerce' model. Competitors like xAI are also eyeing ad-based revenue.
Major AI companies face unprecedented legal risks as courts rule against unauthorized use of copyrighted material in model training. With potential liabilities reaching $75 billion, the industry scrambles to address data sourcing practices while balancing innovation and compliance.
Cohere, a Toronto-based AI firm specializing in enterprise large language models, has secured $500 million in funding led by Nvidia and others. The investment raises its valuation to $6.8 billion as it expands its AI offerings and prepares for potential acquisitions and an IPO.
