OpenAI launched the Responses API, enabling developers to build AI agents for tasks like web searches and file mining. It includes a web search tool, computer interaction capability, and document search feature. Additionally, OpenAI introduced the Agents SDK to coordinate AI agent workflows, enhancing its developer toolkit. The Responses API will replace the Assistants API by mid-2026, incorporating developer feedback.

DeepSeek plans to launch its new AI model, DeepSeek-R2, which features enhanced coding capabilities, improved multilingual reasoning, and reduced operational costs. While the official release date remains unconfirmed, the model is expected to strengthen DeepSeek's position in the global AI market.

Google has integrated Gemini AI into Gmail, enabling users to add events directly to Google Calendar by analyzing email content. A "Add to Calendar" button simplifies the process, enhancing productivity. While convenient, reliance on AI poses risks of misinterpretation or errors, requiring user confirmation for accuracy. The feature is rolling out in phases to various groups, showcasing Google's AI expertise and focus on improving user experiences.

Poe's 2025 AI market report reveals significant shifts in text, image, and video generation. Google shows uneven performance across modalities, with strong presence in image and video but declining text shares. Black Forest Labs leads in image generation, while video generation sees fierce competition from Runway and Google's Veo-2. New models rapidly erode older versions' usage, challenging companies to adapt their lifecycle strategies. Despite OpenAI and Anthropic's dominance in text AI, new e

AI21 Labs launched Maestro, a platform enhancing large language model outputs by automating error reduction and optimizing responses. It simplifies complex prompts, improves accuracy by up to 50%, and streamlines enterprise AI workflows. Maestro is in early access, with broader availability planned for later this year.

DouBao's COMET system optimizes communication for MoE models in distributed training by reducing overhead and improving efficiency. It introduces shared tensor dependency parsing and adaptive load balancing to align computation with communication, dynamically adjust resources, and minimize GPU idling. Tested on an 8-GPU cluster, COMET reduces forward latency by 31.8%-44.4% compared to baselines, with speedups of 1.28x-2.37x on MoE layers. Deployed across thousands of GPUs, it saves millions of G

OpenAI signed a $11.9 billion, five-year deal with CoreWeave, including $350 million in equity acquisition. This reduces investor concerns about CoreWeave's reliance on Microsoft, previously 62% of its revenue. Backed by NVIDIA, CoreWeave operates an AI-focused cloud network with over 250,000 GPUs. The deal strengthens OpenAI's computing power and shareholding in CoreWeave amid growing tensions with Microsoft, which is developing its own AI models. CoreWeave plans to use IPO funds to reduce its

Qualcomm Technologies is acquiring Edge Impulse, a startup providing a cloud platform for developers to train AI models optimized for connected devices. Edge Impulse's tools automate data collection, simplify model training, and offer deployment to multiple hardware platforms. The acquisition follows Qualcomm's recent launch of Dragonwing, a processor lineup for connected devices, with plans to expand Edge Impulse's support for more Dragonwing processors.

Sony is developing an AI-driven prototype featuring Aloy from "Horizon: Forbidden West." The demo showcases voice-based conversations using AI-generated speech and facial expressions, leveraging technologies like OpenAI's Whisper, GPT-4, Llama 3, and Sony's EVS and Mockingbird systems. Though demonstrated on PC, some components work on PS5 with minimal overhead. This remains a Guerrilla Games collaboration for internal tech demonstrations, raising questions about its broader implications for gam

In the 2025 AI competition, Alibaba's QWQ-32B, DeepSeek R1, and OpenAI's O1 Mini lead in reasoning and efficiency. QWQ-32B excels with 32B parameters, offering powerful reasoning via reinforcement learning. DeepSeek R1, with 671B parameters, is a reasoning leader but activates only 37B at once for efficiency. O1 Mini provides cost-effective STEM solutions. The trend shifts toward smaller, optimized models balancing performance and computational costs.

Microsoft is celebrating its 50th anniversary with an event on April 4th featuring a retrospective for employees and a focus on the future of its Copilot AI assistant. Rumors suggest Microsoft may reveal progress on its proprietary AI models, codenamed MAI, potentially competing with OpenAI's offerings. The event will include CEO Satya Nadella and AI head Mustafa Suleyman, highlighting AI as key to Microsoft's next 50 years.

Foxconn unveiled FoxBrain, a large language model for enhancing manufacturing and supply chain management. Built on Meta's Llama 3.1 in four weeks using 120 Nvidia H100 GPUs, it supports reasoning, analysis, computation, and code generation. Foxconn plans to open-source FoxBrain, promoting industry collaboration while advancing AI-driven efficiency and innovation.

Google renamed the Gemini extension to the Gemini app in the beta version of its Android platform. It now uses Gemini Flash 2.0, a local AI model enhancing performance, reasoning, and efficiency. This update advances Gemini's integration and intelligence.

Text2Data is a novel technique by Salesforce AI Research that enhances text-to-data generation in low-resource settings. It uses a two-phase diffusion-based approach: first learning data distribution from unlabeled data, then fine-tuning with constrained optimization on limited text-annotated data. This prevents catastrophic forgetting and improves controllability, achieving superior generation quality across modalities. The method optimizes three key components for balanced learning of data dis

Gree Electric has integrated its multi-model framework with DeepSeek R1, enhancing AI capabilities in voice-controlled air conditioners. This advancement enables seamless connectivity with smart home devices, offering users a more intelligent experience. Gree plans to develop specialized AI models for smart homes, focusing on energy management, air quality, security, and lighting control. The company's rebranding to "Dong Mingzhu Health Home" aims to leverage trust and quality awareness, showcas

Sony Music is cracking down on AI-generated counterfeit music, removing over 75,000 tracks featuring top artists. While this highlights their commitment to copyright protection, concerns remain about AI's broader impact. Experts warn that free AI-generated music threatens legitimate business models. A UK proposal allowing AI companies to use content without explicit permission has sparked industry opposition, as it may harm music industry interests and competitiveness. Sony Music advocates for m

Google launched Gemini Embedding, a new experimental text embedding model via its Gemini developer API. This model transforms text into numerical representations capturing semantic meaning, useful for document retrieval and classification. Trained on the Gemini AI series, it surpasses Google's previous model (text-embedding-004) in performance, supporting larger text inputs, code, and over 100 languages. Still in the experimental stage, Google plans to release a stable version soon.

Elon Musk's legal battle against OpenAI was partially denied, but a federal judge raised concerns about OpenAI's transition from a nonprofit to a for-profit entity. Judge Yvonne Gonzalez Rogers warned of potential "irreparable harm" using public funds for this shift and questioned the plans. Musk's team plans to proceed with an expedited trial in fall 2025, while regulatory scrutiny intensifies over OpenAI's restructuring. Evidence showed Musk knew of OpenAI's possible profit-driven shift, weake

Apple has postponed the launch of new Siri features planned for iOS 18, citing infrastructure challenges and unresolved bugs. The development team struggled with a dual-system approach due to tight deadlines, leading to complexity and performance issues. Executives raised concerns about functionality and promotional readiness, while some employees criticized AI leadership, warning of continued lag in the field without changes. This delay highlights technical and management challenges, prompting

Creatopy, an AI-driven platform for automating digital ad creation, appoints Tammy Nam as its new CEO. Nam, with prior leadership roles at PicsArt and Viki, joins Creatopy's board while former CEO Dan Oros moves to an advisory role. Backed by a $10 million Series A funding round, Creatopy reports 400% revenue growth in the past year, securing clients like AstraZeneca and NASCAR. The platform stands out with its intuitive interface, brand safety features, and high-touch customer service complemen