AI 日报 — 2026-06-28

OpenAI and Broadcom have collaborated to launch Jalapeño, a new custom AI chip. This chip is specifically designed to optimize large language model (LLM) inference. The goal of Jalapeño is to enhance the performance, efficiency, and scalability of AI systems, addressing key areas for advancement in artificial intelligence.

NVIDIA NeMo AutoModel significantly accelerates fine-tuning Mixture-of-Experts (MoE) models by building on HuggingFace Transformers v5. It integrates Expert Parallelism, DeepEP fused all-to-all dispatch, and TransformerEngine kernels, leveraging v5's dynamic weight loading. This results in 3.4-3.7x higher training throughput and 29-32% less GPU memory compared to native Transformers v5, using the same API. NeMo AutoModel enables efficient scaling of MoE models, even for frontier-scale models where v5 runs out of memory.

PP-OCRv6, PaddleOCR's latest universal OCR model, offers three tiers (1.5M to 34.5M parameters) for diverse deployment. The small and medium tiers support 50 languages. It achieves 86.2% detection Hmean and 83.2% recognition accuracy with the medium tier, improving upon PP-OCRv5. Key advancements include PPLCNetV4 backbone, RepLKFPN for detection, and EncoderWithLightSVTR for recognition, ensuring accurate, structured text output across various real-world scenarios.

New research in "Nature" highlights AMIE, an AI system, for its potential in long-term health condition management. Utilizing Gemini models, AMIE features an empathetic dialogue agent and a reasoning agent that cross-references clinical knowledge. In a blinded study with patient actors, AMIE matched physicians in management reasoning and surpassed them in plan preciseness and guideline alignment, suggesting its future role in supporting medical care and freeing up physician time. Further studies are exploring its use in clinical settings.

OpenAI research reveals AI agents are revolutionizing work by facilitating longer, more intricate tasks. This advancement significantly boosts productivity across various job functions, demonstrating a transformative impact on the modern workplace.

The FFASR Leaderboard, a collaboration between Treble Technologies and Hugging Face, is the first open, community-driven benchmark for evaluating ASR models in realistic far-field acoustic conditions. It addresses the significant gap between clean-speech benchmarks and real-world performance, where reverberation and noise degrade accuracy. The leaderboard uses hybrid wave-based simulation across 14 diverse rooms, validated against real measurements, to assess models' robustness. It tracks Word Error Rate (WER) and RTFx, encouraging development of models suited for complex environments like AI voice agents and in-car assistants.

Immunologist Derya Unutmaz utilized GPT-5 Pro to unravel a three-year-old mystery concerning T cell behavior. This significant advancement holds promise for future research in both cancer and autoimmune diseases. The AI's contribution provided crucial insights, potentially accelerating scientific understanding and therapeutic development in these critical areas of medicine.

Hugging Face now ships weekly updates for `huggingface_hub`, its core Python client, using an automated workflow. This system, built with open-source tools like GitHub Actions and an open-weights AI model (GLM-5.2), automates mechanical tasks like version bumping and publishing. A human remains in the loop to review and edit AI-drafted release notes and announcements, ensuring accuracy and quality. This approach aims for reusability, allowing other maintainers to adapt the workflow.

Google Search tools are revolutionizing thrift and vintage shopping, with "vintage" and "how to thrift" trending. AI Mode helps plan shopping trips, even suggesting gluten-free brunch spots. Google Lens uncovers hidden gems by identifying items and their value. Circle to Search finds similar items online from images. Virtual Try-On lets users digitally preview vintage clothes. Additionally, Lens can help users assess the resale value of their own items, promoting a circular economy.

Google I/O 2026 leveraged AI tools like Gemini and Nano Banana to create various elements, from a "TPU Training Day" film blending human artistry with experimental technology, to the visual brand identity. AI also powered immersive pre-show experiences, including "Jellectronica," which translated jellyfish movements into music, and "Infinite Scaler," a video game where players generated levels using AI. This showcased how AI can enhance creativity and efficiency in event production.

OpenAI has launched Patch the Planet, a new Daybreak initiative. This program aims to assist open-source maintainers in identifying, validating, and resolving security vulnerabilities. It leverages a combination of artificial intelligence and expert review to support these efforts.

Google's I/O 2026 announcements are highlighted in a new quiz, "vibe coded" in Google AI Studio. This quiz, built by an editor with no coding experience using Gemini and Google AI Studio, demonstrates how the Antigravity coding agent allows users to create with the latest Gemini models. The quiz showcases new applications for Google AI Studio, encouraging developers and builders to explore its capabilities.

Transformers.js allows Web developers to integrate AI models. A challenge arises with caching: even if multiple applications use the same AI models or Wasm runtime files, browsers re-download and re-cache them if they originate from different top-level sites. This is due to cache isolation, a security measure preventing timing attacks by partitioning caches based on the top-level site and current-frame site, leading to redundant downloads and storage.

Google Finance has released new updates, including an Android app, to help users track investments and stay informed. Users can now consolidate portfolios, gain insights into asset allocation, and utilize a research tool. The platform also offers customized market intel briefings based on user preferences. The new Android app provides real-time data, news, and an AI research tool, with an iOS app planned for later this year.

Users can launch a private, OpenAI-compatible LLM endpoint on Hugging Face infrastructure with a single command. This allows for quick setup of models for testing, evaluations, or batch generation, with billing per-second for hardware usage. The endpoint is gated and requires an HF token for access, ensuring privacy. Users can query the server from various platforms and scale to larger models by adjusting hardware flavors and parallelization settings.

Google is investing $1.5 billion to expand its Jackson County, Alabama data center by 2027, funding its own power and infrastructure. This expansion includes a $2 million Energy Impact Fund with TVA and CAANEAL for local energy efficiency. Google is also donating $550,000 for STEM kits for students, building on its existing community efforts like water stewardship, digital skills training for over 130,000 Alabamians, and job creation.

OpenAI is actively involved in establishing shared standards for advanced AI. They contribute to this effort by supporting the development of evaluation frameworks and safety practices. Furthermore, OpenAI promotes global cooperation in the AI field through its involvement with the Appia Foundation, aiming to foster a unified approach to AI development and deployment.

Google is expanding its commitment to Virginia with new community investments. These initiatives aim to create thousands of local jobs and prepare the future workforce by funding electrical apprenticeship training, targeting 2,741 additional apprentices by 2030. Furthermore, Google is launching a $15 million Energy Impact Fund to reduce utility bills for Virginians through home repairs and energy-efficiency upgrades, alongside investing in over 500 megawatts of new energy capacity.

OpenAI has unveiled a preview of GPT-5.6 Sol, their next-generation model. This new iteration promises enhanced capabilities across several key domains, including coding, scientific research, and cybersecurity. A significant feature of GPT-5.6 Sol is its integration with OpenAI's most advanced safety stack, suggesting a strong focus on secure and responsible AI development.

Omio is leveraging OpenAI to revolutionize travel experiences, focusing on conversational AI. This integration is not only accelerating their product development but also fundamentally transforming Omio into an AI-native company. Their strategy centers on using AI to enhance user interaction and streamline their operational processes.