Pulse

Cohere has announced a new partnership with the government of Quebec. This collaboration aims to leverage Cohere's AI technology within the Quebec region. Further details on the specific applications or goals of this partnership were not immediately available. AI

Together AI has significantly expanded its cloud computing resources, adding thousands of new chips including NVIDIA's B200 and B300 accelerators. This move is aimed at bolstering their dedicated model inference services, providing enhanced capabilities for AI model deployment and operation. AI

Unitree, a Chinese robotics company, is poised to lead the global market due to its rapid iteration cycle. This accelerated development is expected to drive significant advancements in next-generation robotics. The company's fast-paced approach suggests a strong competitive advantage in the evolving robotics landscape. AI

Perplexity has released research detailing the impact of its autonomous agent, Computer, on knowledge work. The study, conducted in collaboration with Harvard, found that Computer significantly reduces task completion time and cost compared to traditional search methods. Users reported higher satisfaction and increased autonomy when using Computer, which handles complex, multi-field queries that often extend beyond typical search capabilities. AI

Unitree, a robotics company, has been added to the U.S. Department of Defense's Section 1260H list of Chinese military companies. This designation places Unitree alongside other prominent Chinese tech firms like BYD, Alibaba, Baidu, and Tencent. SemiAnalysis predicts that despite this, Unitree is poised to dominate the global robotics market due to favorable timing. AI

Researchers are exploring novel approaches to enhance the efficiency and effectiveness of attention mechanisms in transformers. Several papers introduce methods to mitigate issues like over-smoothing and computational bottlenecks, particularly in graph transformers and large language models. Techniques include capacity-controlled attention gating, analyzing attention sinks to differentiate between adaptive no-op and broadcast mechanisms, and developing sparse attention strategies for ultra-long contexts. These advancements aim to improve model performance on various benchmarks while reducing computational costs. AI

NVIDIA has open-sourced parts of its cuDNN library, a significant move after 12 years of it being closed-source. This release includes over 20 Mixture-of-Experts (MoE) kernels and NSA sparse attention kernels. The codebase for these kernels is largely written in Python CuTe-DSL, with public documentation now available. AI

Google DeepMind has introduced an AI co-clinician research initiative aimed at assisting healthcare professionals and patients. This system utilizes live video and audio to analyze physical symptoms in real-time, such as a patient's gait or breathing. In testing, the AI demonstrated strong performance, matching or exceeding physicians in 68 out of 140 assessed areas, including triage, and made zero critical errors in 97 out of 98 primary care queries under the NOHARM safety framework. AI

Alibaba's Qwen team has released FlashQLA, a new set of high-performance linear attention kernels developed using TileLang. These kernels are designed to improve the efficiency of attention mechanisms in large language models. The team also shared benchmark results for their Qwen models, showcasing performance across various configurations. AI

Google DeepMind's "Experience AI" program has trained over 30,000 educators globally since 2023, with 93% reporting increased AI knowledge and 87% feeling more confident teaching the subject. The initiative, developed with Raspberry Pi, has reached 2.9 million students across 180 countries in 19 languages. The program is now expanding into Latin America with $4.6 million in funding from Google.org, aiming to train an additional 24,000 educators and impact 1.25 million students by 2028. AI

Google DeepMind has introduced Decoupled DiLoCo, a novel approach to training advanced AI models that enhances resilience and flexibility across data centers. This system can train models like Google's 12B Gemma model across geographically dispersed regions using low-bandwidth networks and can even mix different generations of hardware, such as TPU6e and TPUv5p. Decoupled DiLoCo is designed to be self-healing, isolating and continuing training through artificial hardware failures and reintegrating units when they come back online, addressing the synchronization issues that typically stall AI training. AI

Runway has released several updates to its video generation platform. Seedance 2.0 is now available in 1080p, via the iOS app, and through the Runway API. Additionally, users can now animate Runway Characters using scripts, bringing them to life with text prompts. AI

Google AI has released Gemini 3.1 TTS and Gemini 3.1 Flash TTS, their newest text-to-speech models. These models offer enhanced expressiveness and control, introducing audio tags to guide vocal style, pace, and delivery through natural language commands. The audio tags are designed to be an intuitive way for users to shape the output of the text-to-speech models. AI

Alibaba has released its Qwen3.6-27B model, an open-source, dense model that demonstrates strong coding performance, outperforming a significantly larger predecessor on key benchmarks. This new model is natively multimodal, capable of processing both vision and language inputs. The release has been accompanied by rapid integration with popular AI tools like vLLM and SGLang, enabling local execution and broader accessibility. AI

Researchers are developing advanced agent frameworks to improve AI reliability and efficiency across various domains. Google introduced an agentic RAG system that enhances enterprise query handling by iteratively searching for complete context, boosting accuracy by up to 34%. Hugging Face demonstrated a multi-agent economy simulation using a small 3B model, highlighting the trade-offs between model size and real-time performance. Other research explores methods for reliable tool use, regulatory compliance through agent-to-agent protocols, dynamic benchmarking for agent behavior, and robust self-evolution mechanisms for AI agents. AI

Researchers have developed several new methods to accelerate large language model (LLM) inference through speculative decoding. AdaPLD improves retrieval and draft construction by using semantic similarity and branched hypotheses, achieving up to 3.10x speedup. SSSD combines n-gram matching with hardware-aware speculation for up to 2.9x latency reduction without training. D^2SD uses a dual diffusion model and confidence-guided prefix trees to enhance acceptance rates, while TAPS optimizes prefix tree selection for diffusion-drafted decoding, yielding up to 7.9x speedup. KnapSpec treats draft model selection as a knapsack problem to maximize throughput, achieving up to 1.47x speedup, and Vegas uses verification-guided sparse attention for improved decoding throughput. Additionally, LK Losses directly optimize the acceptance rate during training, leading to gains of 8-10% in average acceptance length. AI

Google Research has introduced ReasoningBank, a novel framework designed to enhance AI agents' ability to learn from their experiences, both successes and failures, after deployment. This system distills generalizable reasoning strategies from past interactions, allowing agents to continuously improve and avoid repeating mistakes. Separately, new research explores optimizing multi-agent communication through latent representations and introduces Agent Evolving Learning (AEL) for agents operating in open-ended environments, focusing on how to effectively use remembered information. Additionally, DeepSeek has released preview models of its V4 series, offering large context windows and advanced capabilities at a significantly lower cost than comparable frontier models. AI

OpenAI has published a series of research papers detailing advancements in reinforcement learning. These include achieving superhuman performance in Dota 2 with OpenAI Five, developing benchmarks for safe exploration in RL, and quantifying generalization capabilities with the CoinRun environment. The company also explored novel methods like prediction-based rewards for curiosity-driven exploration, learning policy representations in multiagent systems, and an experimental metalearning approach called Evolved Policy Gradients for faster training on new tasks. Further research addresses variance reduction in policy gradients and the equivalence between policy gradients and soft Q-learning, alongside challenging robotics environments for multi-goal RL. AI