General Motors appoints Barak Turovsky as its first Chief Artificial Intelligence Officer, signaling a major push for AI integration in vehicles and operations. Turovsky, with prior roles at Google and Cisco, will lead GM's AI efforts, focusing on EVs, autonomous driving, and process optimization. Reporting to SVP Dave Richardson, his team aims to enhance product performance and customer experience, reinforcing GM's commitment to digital transformation and innovation in mobility solutions.

Google updated its Gemini app for iOS/iPadOS with six new lock screen widgets to enhance AI interaction. These widgets offer quick access to features like text prompts, live conversations, and file sharing. The move aims to attract iPhone users away from Siri or ChatGPT. Apple's Siri upgrade remains slow, providing Google an opportunity to promote Gemini and build user familiarity. Widgets include text input, voice activation, camera use, and image/file sharing capabilities.

Microsoft unveiled Dragon Copilot, an AI system for healthcare that generates clinical notes using voice input and ambient listening tech from Nuance. It offers multilingual note-taking, automates tasks like summaries and referrals, and aims to reduce administrative burdens for healthcare professionals. Microsoft highlights responsible AI design with safeguards for accuracy and compliance. Other tech firms like Google Cloud also provide AI tools for healthcare, while the FDA warns of risks like

Ubtech successfully conducted the world's first multi-unit, multi-scenario collaborative training for humanoid robots at Zeekr's 5G factory. Using swarm intelligence and BrainNet software, the Walker S1 robots demonstrated multi-task capabilities in complex industrial environments, advancing from single-unit autonomy to collective intelligence. This breakthrough paves the way for widespread industrial adoption of humanoid robots.

ByteDance launched the domestic version of Trae, China's first AI-native IDE. It supports switching between DeepSeek R1 and V3 models, features code completion, bug fixing, and natural language-based code generation. A new Builder mode enables project framework creation via natural language, lowering development barriers. Optimized for Chinese developers, future updates will include model customization options. Available on Mac and Windows, with a Linux version coming soon.

DeepSeek, a Hangzhou-based AI startup, revealed financial data showing its AI models, DeepSeek-V3 and R1, could achieve a cost-to-profit ratio of up to 545% under ideal conditions. Daily operational costs are $87,072, with potential revenues of $562,027, highlighting an optimized cost structure compared to rivals. The company also offers discounted API pricing during off-peak hours, aiming to attract more users and investors while competing globally.

iFlytek's Spark X1 has been upgraded, enhancing mathematical capabilities and matching industry leaders despite fewer parameters. A new medical model, Spark Medical X1, outperforms GPT-4 in healthcare tasks. The update also improves education tools, introduces joint machines with Huawei for customized models, and launches domain-specific solutions for various industries. Additionally, the new dual-engine AI Learning Machine T30 series combines Spark X1 and DeepSeek for personalized tutoring in e

OpenAI plans to integrate its AI video tool Sora into ChatGPT, though no timeline has been set. Sora, currently a standalone web app for generating 20-second videos, may see limited features in ChatGPT. OpenAI aims to expand Sora's platform reach, explore mobile apps, and enhance its image generation capabilities with a new Sora-powered AI tool, potentially creating more photorealistic images. A new version of Sora Turbo is also in development.

Apple is reinventing Siri for the generative AI era. A modernized, conversational version will debut with iOS 20 in 2027. Before that, an updated "dual-brain" Siri, integrating traditional commands and advanced queries using user data, will launch in spring 2026. This "LLM Siri" will first be previewed at WWDC in June. Advanced features may follow in 2028.

Google's AI chatbot Gemini is rolling out the "Save Information" feature to some free users. Previously exclusive to paid subscribers, this function lets users input personal preferences for customized responses, currently in English only. Users can manage settings via gemini.google.com/saved-info or during chats, and toggle the feature on/off in settings. This update differs from the "Recall" feature, which remains exclusive to Gemini Advanced for referencing past chat history.

Lenovo unveiled AI Display and AI Stick at MWC 2025, featuring a 32 TOPS NPU for local AI processing via USB. These accessories boost AI computing power without relying on host devices' resources, potentially expanding AI application scope. However, their commercialization remains uncertain.

Baidu announces the launch of Wenxin 4.5 on March 16, marking the second anniversary of its generative AI product. The new model features improved foundational capabilities, multimodal processing, and deep reasoning, expanding its application in complex tasks. This release highlights Baidu's ongoing AI innovation.

Pika Labs launched Pika 2.2, featuring 10-second video generation, 1080p resolution, and "Pikaframes" for smooth transitions. It supports cinematic ratios and delivers high-quality output for professional use. Available at pika.art.

DeepSeek has open-sourced 3FS, a high-performance parallel file system designed for modern SSDs and RDMA networks. It delivers up to 6.6 TiB/s read throughput in an 180-node cluster and 3.66 TiB/min in benchmark tests. With a decentralized architecture ensuring strong consistency, 3FS is optimized for AI workloads like training and inference. Additionally, DeepSeek released Smallpond, a lightweight data processing framework leveraging 3FS for PB-scale datasets.

OpenAI CEO Sam Altman reveals GPT-4.5's phased rollout due to GPU shortages. The model, costing $75 per million input tokens and $150 per million output tokens, will first reach ChatGPT Pro users, then Plus users. OpenAI plans to add tens of thousands of GPUs and is exploring custom AI chips and data centers to address computing power limitations.

Tencent launched HunYuan Turbo S, a fast-thinking model with instant responsiveness, doubling typing speed and reducing latency by 44%. It combines fast and slow thinking for improved performance in knowledge, reasoning, and content creation. Using a Hybrid-Mamba-Transformer architecture, it lowers costs and enhances efficiency. Available via Tencent Cloud API with a free trial, pricing is reduced to 0.8 yuan per million input tokens and 2 yuan per output tokens.

Figure, a Bay Area robotics startup, plans to accelerate its home robot testing timeline with the "Alpha Test" of its Figure 02 model by late 2025. This move is powered by Helix, its proprietary Vision-Language-Action (VLA) model, which enables robots to process visual and language data simultaneously. Although industrial deployments remain a priority due to higher profitability, Figure aims to eventually address household needs, competing with firms like Tesla and Apptronik. Home testing will f

Inception Labs unveils Mercury Coder, a diffusion-based large language model (dLLM) that processes text sequences simultaneously, achieving speeds up to 10x faster than traditional models. Using a coarse-to-fine generation strategy, it matches or surpasses models like GPT-4o Mini and Claude 3.5 Haiku in some scenarios. Cost-efficient and available via API or local deployment, it partners with Fortune 500 companies to reduce AI costs. This novel approach challenges conventional text generation me

Meta is developing a standalone app for its AI chatbot, set to launch in Q2, competing with ChatGPT, Gemini, and Copilot. Currently integrated into Facebook, Instagram, and other platforms, the new app aims to attract non-users. Features include question-answering, image generation, and personalized suggestions via a memory function. Meta plans to invest $65 billion in AI research and will host an AI-focused event on April 29. OpenAI's CEO humorously suggested developing a social platform in res

Microsoft launched the macOS version of Copilot, featuring dark mode and Command + Space shortcut activation. The app allows users to upload images and generate content via a web interface. It's now available in the US, UK, and Canada, with an iPad update adding split-screen support. Users can log in with Apple ID on iOS devices and upload files for summaries. Unlimited access to Copilot Voice and Think Deeper features is now free for all users.