Modern RL-based post-training for large language models (LLMs) co-locate trajectory sampling and policy optimisation on the same GPU cluster, forcing the system to switch between inference and training workloads. This serial context switching violates the single-program-multiple-data (SPMD) assumption underlying today's distributed training systems. We present Echo, the RL system that cleanly decouples these two phases across heterogeneous "inference" and "training" swarms while preserving statistical efficiency. Echo introduces two lightweight synchronization protocols: a sequential pull mode that refreshes policy weights according to API call for minimal bias, and an asynchronous push-pull mode that streams version-tagged rollouts through a replay buffer to maximise hardware utilisation. Training four representative RL workloads with Qwen3-4B, Qwen2.5-7B, Qwen3-30B-A3B-Thinking-2507 and Qwen3-32B on a geographically distributed cluster, Echo matches a fully co-located Verl baseline in convergence speed and final reward while off-loading trajectory generation to commodity edge hardware. These promising results demonstrate that large-scale RL for LLMs could achieve datacentre-grade performance using decentralised, heterogeneous resources.

Researchers in California have achieved a significant breakthrough with an AI-powered system that restores natural speech to paralyzed individuals in real time, using their own voices, specifically demonstrated in a clinical trial participant who is severely paralyzed and cannot speak. This innovative technology, developed by teams at UC Berkeley and UC San Francisco, combines brain-computer interfaces (BCI) with advanced artificial intelligence to decode neural activity into audible speech. Compared to other recent attempts to create speech from brain signals, this new system is a major advancement. GET SECURITY ALERTS & EXPERT TECH TIPS – SIGN UP FOR KURT'S'THE CYBERGUY REPORT' NOW The system uses devices such as high-density electrode arrays that record neural activity directly from the brain's surface. It also works with microelectrodes that penetrate the brain's surface and non-invasive surface electromyography sensors placed on the face to measure muscle activity.

We present a numerically robust, computationally efficient approach for non-I.I.D. data stream sampling in federated client systems, where resources are limited and labeled data for local model adaptation is sparse and expensive. The proposed method identifies relevant stream observations to optimize the underlying client model, given a local labeling budget, and performs instantaneous labeling decisions without relying on any memory buffering strategies. Our experiments show enhanced training batch diversity and an improved numerical robustness of the proposal compared to existing strategies over large-scale data streams, making our approach an effective and convenient solution in FL environments.

Large language models~(LLMs) have demonstrated impressive performance in various applications, among which role-playing language agents (RPLAs) have engaged a broad user base. Now, there is a growing demand for RPLAs that represent Key Opinion Leaders (KOLs), \ie, Internet celebrities who shape the trends and opinions in their domains. However, research in this line remains underexplored. In this paper, we hence introduce MINDECHO, a comprehensive framework for the development and evaluation of KOL RPLAs. MINDECHO collects KOL data from Internet video transcripts in various professional fields, and synthesizes their conversations leveraging GPT-4. Then, the conversations and the transcripts are used for individualized model training and inference-time retrieval, respectively. Our evaluation covers both general dimensions (\ie, knowledge and tones) and fan-centric dimensions for KOLs. Extensive experiments validate the effectiveness of MINDECHO in developing and evaluating KOL RPLAs.

Recent approaches for semantic correspondence have focused on obtaining high-quality correspondences using a complicated network, refining the ambiguous or noisy matching points. Despite their performance improvements, they remain constrained by the limited training pairs due to costly point-level annotations. This paper proposes a simple yet effective method that performs training with unlabeled pairs to complement both limited image pairs and sparse point pairs, requiring neither extra labeled keypoints nor trainable modules. We fundamentally extend the data quantity and variety by augmenting new unannotated pairs not primitively provided as training pairs in benchmarks. Using a simple teacher-student framework, we offer reliable pseudo correspondences to the student network via machine supervision. Finally, the performance of our network is steadily improved by the proposed iterative training, putting back the student as a teacher to generate refined labels and train a new student repeatedly. Our models outperform the milestone baselines, including state-of-the-art methods on semantic correspondence benchmarks.

The Chinese language is difficult, and perhaps no one has struggled more with it than the inkers and bearers of America's Chinese-character tattoos. Most infamous was probably the tattoo on Britney Spears' hip, which intended to be the character for "mysterious," but ended expressing something closer to "strange." Another popular choice is the Chinese character for "freedom," which mistranslates to mian fei, or "free of charge." I've also seen tattoos intended to represent the Chinese character for "power" represented as dian, which means "electricity" rather than "strength." I got my first tattoo in 2014 at My Tattoo in Alhambra, a road map of Los Angeles in black and red. My second came from a tattoo parlor in a neon lit alley in Shihlin Night Market in Taipei, a Chinese family stamp that depicts the meaning of my last name, a bear.

Artificial intelligence (AI)-powered physiological computing looks at technology that can help us listen to our bodily functions and psychological needs. Dr Youngjun Cho is a world leader in this area of research, which starts with physiological sensing. This includes cardiovascular, respiratory, cortical, perspiratory or pupillary pattern measurements. For example, heart rate monitoring is one of the most powerful features in wearable smartwatches or fitness trackers. With AI and computer vision technologies, such physiological activities can also be measured without wearable devices.

While these companies take different tacks, both have the potential to catalyze tomorrow's advances in drug discovery, new materials, remedies to climate change, closer analysis of military-drone footage and more. And they are hardly the only ones: Microsoft Corp., Amazon.com Inc., Oracle Corp., International Business Machines Corp. and others are also in the AI marathon. The Morning Download delivers daily insights and news on business technology from the CIO Journal team. Consumers and investors, more focused on spasms in the stock market, may not be paying attention to projects not directly connected to lines of business or quarterly results. But research and development often hatch products that vault beyond a lab's original aims.

English based datasets are commonly available from Kaggle, GitHub, or recently published papers. Although benchmark tests with English datasets are sufficient to show off the performances of new models and methods, still a researcher need to train and validate the models on Korean based datasets to produce a technology or product, suitable for Korean processing. This paper introduces 15 popular Korean based NLP datasets with summarized details such as volume, license, repositories, and other research results inspired by the datasets. Also, I provide high-resolution instructions with sample or statistics of datasets. The main characteristics of datasets are presented on a single table to provide a rapid summarization of datasets for researchers.

Samsung announced its first Galaxy S smartphone in the heady days of 2010, and at the time, people were too jazzed by its 4-inch Super AMOLED screen and 1GHz processor to fret much about its cameras. The same could be said of Samsung itself -- the company's original US press release mentioned them a grand total of zero times outside of the spec sheet. Eleven years and millions of Galaxy phones later, cameras have become a crucial part of Samsung's smartphone identity. If you needed any proof, just look at the company's new flagship devices, which go on sale today: the Galaxy S21 and S21 Plus pack a total of four cameras each, while the high-end S21 Ultra sports five and a more pronounced focus on telephoto shooting. And while pundits and reviewers tend to go back and forth on the merits of Samsung's approach to cameras, most of them (myself included) were impressed with what the company pulled off this year.