Hugging Face Daily PapersHy-Nomas is an 8B parameter diffusion model that supports free-form text instructions and camera control across multi-round interactions. It follows multi-round text and camera instructions by combining causal attention with structured video descriptions, outperforming existing world models in human evaluations. The model weights, code, and benchmarks are open-sourced.
Oracle is establishing a diverse AI computing power portfolio, with the SambaNova SN50 RDU chip to be added to OCI cloud services. This portfolio also includes NVIDIA, AMD chips, Positron Atlas chips, Google TPUs, and Cerebras wafer-level chips, with future support for Intel Diamond Rapids processors.
Browser UseCloudflare often blocks standard headless Chrome due to automated features (navigator.webdriver is true, Playwright's default headless shell user agent does not display HeadlessChrome), and it usually starts from cloud server IP without cookies.
Ollama releasesLiquid AI releases the LFM3 small model family, including the 1.7B parameter lfm3 text model (256K token context), lfm3-vl supporting image understanding, lfm3-audio with 18 language speech processing capabilities, and the general model lfm3-omni for local agents and structured output; the d1 decision model is upgraded to 2B parameters, supporting text, image, and audio decision-making, with some models supporting Apple hardware.
NVIDIA Technical Blog6565NVIDIA CUDA 13.4 released, enabling GPU sharing in Kubernetes through memory-enforced mechanisms, allowing a single GPU to be split into hardware-isolated instances for multiple AI workloads, improving resource utilization.
Together AI Blog3838Together AI has launched an auto-repair feature for GPU clusters, which continuously monitors node health on both virtual machines and host machines. When a failure is detected, it recommends minimizing repair actions and provides evidence, allowing users to choose automatic execution or manual approval. This feature is now fully available, supporting passive health checks, repair recommendations, automatic repair policies (including fault and node-level authorization), idle wait control, upgrade tiers, and safety guards, and is performed via API and console; the preview version provides repair commitment support based on node tags for third-party detection systems.
Qianli Technology has launched an AI 4D millimeter-wave radar, equipped with a 10 TOPS NPU, supporting local AI perception and satellite architecture parallel output, achieving end-to-end AI processing from antenna design to point cloud output. The radar's point cloud density is enhanced 10 times, detection range is increased by 40%, ghost shadow elimination rate reaches 95%, and spatial angular resolution is improved 12 times. It provides three working modes: edge-side, satellite architecture, and compatible, to adapt to different functional levels and platform requirements.
Databricks Blog4545Databricks Vital Skills enhances Genie Code's knowledge in the healthcare and life sciences field through the open-source Agent Skills standard, achieving a score improvement in the BiomniBench benchmark test.
The DecoderDeepMind has released Genie 4, the company's first real-time voice-generating instant world model, where users can move scenes with the keyboard and guide events with text prompts. Genie 4 is built on the video model Gemini Omni and is planned and simulated in the background by the Gemini large model family, supporting background sound effects, character dialogues, and real-time interaction. It is currently only available to selected testers and will be opened to Google AI Ultra subscribers in the future, integrating SynthID watermarking to mark AI-generated content.
NVIDIA Technical BlogNVIDIA RTX Spark is a new Windows PC platform for AI development, creation, and gaming, suitable for thin and light laptops and compact desktops, equipped with NVIDIA Blackwell RTX GPU and Grace CPU, offering up to 128 GB of unified memory, and supporting the full NVIDIA CUDA stack and RTX technology (including ray tracing and DLSS 5). The platform supports game developers in building RTX-based applications.
TechCrunch · AIOpenAI has published solutions to three mathematical problems, including an unexpected refutation of the Exponential Time Hypothesis (ETH). The hypothesis plays a significant role in computer science, affecting the understanding of computational complexity. OpenAI's solutions are based on existing work, sparking a discussion in the scientific community about the norms of AI contributions, with Anthropic subsequently publishing its own ETH rebuttal.
Google Developers Blog3434Google Released HappyDelivery, a Conversational Business Prototype Based on LangGraph and Gemini, Implemented on GCP; Ingress Layer Processes WhatsApp Inputs via GKE/Cloud Run and Cloud Pub/Sub
Databricks BlogDatabricks improves document extraction accuracy by building a document extraction process that includes a review step, using ai_extract combined with JSON schema and global instructions, and leveraging MLflow Prompt Registry for versioned configuration. Finally, it validates the improvement effect on a fixed set of manually labeled data, enhancing the overall performance of the system.
Google Research5858Google Research proposes a sleep framework that enables LLMs to achieve continuous learning through alternating active (awake) and sleep phases; the sleep phase utilizes the MoE and Hope architectures for parameter expansion and knowledge distillation, integrating short-term memory into long-term parameters, supporting the retention of long-context knowledge up to 10M tokens and enhancing complex reasoning capabilities.
AWS Machine Learning Blog3434Exowatt customizes task-specific fine-tuned models Qwen3-1.7B and Qwen3-14B for edge and cloud deployment targets through Amazon SageMaker AI serverless models, achieving accuracy surpassing state-of-the-art models while reducing costs up to 49 times and shortening the model development cycle from months to days.
AWS Machine Learning Blog3434Property Finder discovered that 96.8% of the labels in its evaluation dataset for visual AI workloads migrated to Amazon Bedrock were predictions from a single model rather than human validation, hindering objective comparison of new models. The company built a trustworthy benchmark dataset by independently labeling 356 images with human experts and, after 27 iterative optimizations with evolutionary prompt tuning, reduced costs by 86% while maintaining an 86% accuracy rate.
AWS Machine Learning Blog3434AWS SageMaker AI introduces SOCI-indexed images functionality, removing container image downloads from the critical path, reducing SageMaker AI training and processing task startup times from minutes to seconds. This feature is achieved by pre-indexing AWS Deep Learning Container (DLC) or adding indexing to custom images, requiring no changes to workload definitions. Benchmark tests show it can reduce GPU idle time by 90% and improve efficiency 8-46 times.
NVIDIA Newsroom5858NVIDIA Vera Rubin NVL72 and NVIDIA Vera CPU are now officially available on Oracle Cloud Infrastructure (OCI), with plans to introduce HGX Rubin NVL8 systems to help organizations accelerate cutting-edge model training, improve inference efficiency, and expand agent workloads.
NVIDIA Technical Blog6565Fine-Tuning NVIDIA Nemotron can adapt general models with local data, enabling them to possess specific local knowledge and cultural precision required for certain applications. This technology allows the model to meet local language needs while understanding and executing local rules. NVIDIA Nemotron achieves this functionality through fine-tuning, making it suitable for scenarios requiring high localization capabilities.
Hugging Face BlogHugging Face releases Old Ideas, Novel Problems study, evaluating the reliability of six LLMs (GPT-5.1 to 5.4, Claude Sonnet 4.5 to 4.6) in judging the reliability of research innovation.
Replit BlogReplit Cloud Agents allow any team member to contribute code directly and securely to the core codebase, by testing each change in a running app, passing review feedback, and sharing configurations, engineers maintain control while lowering the contribution barrier for others.
Ollama releasesLiquid AI has released the LFM3 small model family, including the 1.7B parameter text model lfm3 and variants supporting image and audio functions, with a context window of up to 256K tokens; the d1 decision model has been upgraded to 2B parameters, supporting text, image, and audio inference, with Apple hardware already supporting the rc1 version.
NVIDIA Technical BlogSame story · 2 outlets6565Inclusion AI found that decoding is the memory bottleneck when evaluating Ling-3.0-flash with a batch size of 1 on NVIDIA DGX Spark, where MoE computation and output layer projection increase execution time. MoE computation and output layer projection are bottlenecks.
How the accounts differ:NVIDIA Technical Blog states decoding is the memory bottleneck due to MoE computation and output layer projection, while MarkTechPost highlights the 42.5% throughput increase and 3.5x speedup due to new MoE backend, FP8 output layer, and speculative decoding.
NVIDIA BlogC-BRAIN and NVIDIA are using AI to change the pace of dementia research. They are accelerating the development of new therapies by integrating previously unpublished Alzheimer's research data, including unpublished results and controlled proprietary data. C-BRAIN employs a hybrid centralized and federated architecture to protect data privacy, deploying 64 NVIDIA data center GPUs and cloud nodes, and plans to introduce the NVIDIA Blackwell system to support its expansion goals.
The DecoderThore Graepel's team proposes building AI systems by borrowing AlphaGo's strategy, recording known information in a knowledge base and verifying each conclusion to reduce hallucinations. This method requires that reasoning steps must be independently verified before changing the knowledge state, similar to AlphaGo combining predictive networks with system search. Since reasoning can be broken down into small steps, this solution may be more suitable for deploying small models on edge devices and robots.
MarkTechPost6565Mirendil released its first AI model, Fern 0.1, supporting a context window of up to 10M-Tokens, equivalent to approximately 23 MB of code or 500,000 lines, achieving the first complete loading of a Megatron-LM level codebase.
Ant Group inclusionAI · Hugging Face new modelsSame story · 2 outlets5858inclusionAI/Ling-3.1-flash-1019 released, 560B parameter size, 25B active parameters/Token, 1M context window, improved reasoning ability, AA-Briefcase score 1524 Elo, AutomationBench-AA 68%, Artificial Analysis Intelligence Index v4.3.2 44 points. Supports SGLang and vLLM inference.
O'Reilly Radar6262The Harness-Zero project at Peking University trained the behavior of the Qwen3.5-9B framework into a model, improving its average success rate by 21 points on table editing, multi-application tool use, and chemical tasks, surpassing the original framework+model combination.
OpenAI releases GPT-6.1 Terra, standard API input at $1 per million tokens, cached input at $0.05 per million tokens, output at $5 per million tokens, prices about half of GPT-5.6 Terra.
Ai2 BlogAi2 has released a study titled 'Old Ideas, Novel Problems,' evaluating the reliability of six LLMs (GPT-5.1 to 5.4, Claude Sonnet 4.5 to 4.6) in judging the innovativeness of scientific research. The study found that fine-tuning prompts can lead to accuracy differences of over 50%. The research constructed a benchmark dataset containing 154 human-generated and 154 AI-generated scientific ideas, confirming that models have insufficient stability in judgment and require more reliable methods to evaluate new ideas proposed by AI.
Hugging Face Daily Papers6565The Function-Space Policy Optimization (FSPO) framework significantly improves the SDPO method by storing sparse historical next-word distributions as fixed time targets, combining privileged adaptation with function-space time integration, achieving time separation without maintaining a persistent teacher model.
Hugging Face Daily PapersPropose a unified framework to integrate foundational models with runtime components for reliably executing complex digital tasks, introducing dimensions of work persistence and openness, and identifying five core capabilities and optimization methods.
Hugging Face Daily Papers6565The FORK reward mechanism enhances response quality and reduces redundancy by combining necessity credit driven by scoring with factual scoring, while optimizing recall and precision. Compared to traditional rewards based solely on recall, FORK significantly improves precision and shortens response length while maintaining similar scoring coverage, increasing win rates from 10.6% to 68.0% on RubricHub-Medical and from 22.1% to 74.1% on HealthBench.
MarkTechPostFeatured7171Baidu open-sources Unlimited OCR 2, a 3.34B parameter OCR visual language model that can parse hundreds of pages of books, financial reports, and reports in a single request. The weights are released under the MIT license, and the code is on GitHub.
Why it matters: The original text provides a comparison of SR-SWA's sparse attention mechanism with the hundred-page document benchmark, showing how the contextual cost of long document OCR can be flattened.
MarkTechPost5858vLLM Semantic Router team releases d3-ultra, a 26.09B parameter (including 0.46B visual encoder), 27B open-source multimodal decision model with a context length of 131,072 tokens.
Hugging Face Daily Papers3838DynaFlow coordinates sub-agents by having the main agent synthesize executable processes, achieving adaptive multi-agent collaboration. The framework integrates free-form executable process synthesis into the action space of the main agent, where the main agent determines when collaboration is needed and generates programs for specific tasks, coordinating sub-agents through sequential or parallel execution, structured information exchange, and verification.
Ant Group inclusionAI · Hugging Face new models5858SingProbe is a streaming safety probe based on Qwen/Qwen3.8-Flash-Next, which scores query intent, response unsafety, and hallucination risk at each step of generation by reusing the hidden states of the base model, with less than 0.5% overhead during decoding.
