Whispers

A new benchmark reveals that frontier large language models degrade approximately 25% of documents during extended workflows. Separately, a Fields Medal winner has reported that ChatGPT 5.5 Pro is capable of solving complex PhD-level mathematics problems. AI

A blog post details how to use NVIDIA's AIPerf tool to uncover hidden performance issues in LLM deployments. Initial tests with a local model showed excellent baseline performance, but increasing concurrency revealed a dramatic increase in time-to-first-token (TTFT), with 99% of requests failing a 500ms SLO. The analysis highlighted that the bottleneck is not the model's inter-token latency (ITL), which remained stable, but rather the request queuing and prefill phase, suggesting architectural solutions like better queue management or horizontal scaling are needed. AI

The author details the practical implementation of the A2A Protocol, an open standard for agent discovery and task delegation. This second part focuses on the code, outlining the architecture where the orchestrator acts as both a server and a client. It highlights the importance of the orchestrator being an A2A service to receive structured tasks and emit failure events, contrasting this with a simpler client-only script. The project structure and setup for the shared agent and customer-specific orchestrators are also provided. AI

SPS has launched Philips SpeechLive Health, a new system designed to automate medical note-taking. This ambient documentation tool listens to patient encounters and generates clinical notes, freeing up healthcare professionals. The system utilizes AI models specifically trained on healthcare data to ensure accuracy with medical terminology and context, aiming to minimize errors. AI

Papel is a new application designed to make scientific papers more accessible and engaging, akin to a social media platform. It indexes over 2 million papers and uses on-device AI for instant natural language querying of their content. The app aims to enhance research discovery and community interaction with features like personalized recommendations, an AI chatbot, and interactive quizzes, all while prioritizing user privacy through local data processing. AI

Maynooth University in Ireland is offering a fully funded PhD position focused on Music Information Retrieval for Irish Traditional Music. The studentship will involve research in audio signal processing, machine learning, and computational analysis of traditional Irish music. Applicants from computer science, music technology, audio processing, and AI/ML backgrounds are encouraged to apply, with a deadline of May 29, 2026. AI

Interspectral will lead a Swedish research consortium, including Saab, AMEXCI, and Scaleout Systems, to develop AI-driven quality assurance for aerospace and defense additive manufacturing. The project, named TRUSTAM, will employ federated learning to enhance AI models across different production facilities without the need for raw data sharing. This approach aims to improve quality control in a sensitive industry. AI

A café in Stockholm's Vasastan district, named Andon Café, is utilizing an AI agent named Mona to manage its operations. Mona, powered by Google's Gemini model, handles tasks such as contracts, supplier relations, pricing, and even hiring. While human baristas serve customers, the AI agent acts as the manager. AI

NBA The Run, an arcade-style basketball game developed by Play by Play Studios, is set to launch on June 9 for PlayStation 5, Xbox Series X/S, and Steam. The game draws inspiration from the NBA Street series, featuring 3v3 matches and over-the-top action rather than simulation. It includes a AI

A Business meets AI hackathon successfully developed a minimum viable product within five days, with the BVB team winning the technology category. The discussion covers preparation, teamwork under pressure, and the transition to operational use. The podcast episode is available on Apple Podcasts and Spotify. AI

The Department of Homeland Security is planning an experiment this fall to test autonomous drones and vehicles along the US-Canada border. This joint exercise with Defense Research and Development Canada, named ACE-CASPER, will evaluate the ability of these systems to stream surveillance data across the border using 5G networks. While framed as a public safety and emergency response simulation, the experiment also aims to demonstrate capabilities for gathering real-time battlefield intelligence, using terminology from the Department of Defense. AI

Mythos, a new AI model, has been introduced, described as "The Day AI Sat on Bedrock." The announcement was made on Medium, with further details available via a link to the platform. AI

A new report indicates that no single AI model consistently leads across all benchmarks, with different models excelling in specific areas like coding or math. The evaluation process itself is also complex, as multiple frontier models provide divergent reasoning for their scores when judging agent performance. This suggests that developers need to employ continuous, multi-model evaluation strategies rather than relying on a single leaderboard for model selection. AI

This article highlights five foundational AI papers that are considered essential reading for AI engineers. It aims to explain the core contributions of each paper and their lasting significance in the field. The selection focuses on works that have fundamentally shaped current AI development and understanding. AI

A software engineer utilized Anthropic's Claude Opus model to analyze years of his family's medical records, identifying eleven potential errors or missed opportunities. The system, built as a personal project, fed a comprehensive JSON document of patient data into Claude Opus, which then flagged issues such as drug contraindications, a missing routine test, and a mislabeled prescription. This experiment suggests that LLMs can already outperform existing healthcare systems in specific analytical tasks related to medical record review. AI

A new study published on arXiv investigates emergent misalignment (EM) in large language models, finding it is not a universal phenomenon but rather an artifact of overtraining. Researchers tested 12 open-source models across four families and discovered that EM is more prevalent in larger models and emerges late in the training process. The study suggests practical mitigation strategies, such as early stopping during fine-tuning, which can eliminate EM while retaining most task performance. AI

Frontier AI safety tests might inadvertently create the risks they aim to prevent. Researchers are exploring how these tests could potentially generate or exacerbate the very dangers they are designed to mitigate. This raises concerns about the effectiveness and potential unintended consequences of current AI safety methodologies. Further investigation is needed to understand and address these emergent risks. AI

Databricks has announced the general availability of new features within its Unity Catalog designed to enhance data protection and governance. These capabilities include Attribute-Based Access Control (ABAC) for row filtering and column masking, standardized data classification through governed tags, and automated data detection and tagging. The aim is to provide scalable, consistent, and real-time protection for sensitive data across an organization's entire data estate, reducing manual overhead and improving compliance. AI

This article argues that fine-tuning a vision-language model (VLM) is less about the technical training process and more about strategic decisions made beforehand. The author highlights four key choices that significantly impact the outcome of fine-tuning, suggesting that focusing on these decisions yields better results than solely optimizing training parameters. AI

Pakistan is preparing to issue its first AI

Palantir announced it is adding AI capabilities to its defense and intelligence software. This move aims to enhance the company's offerings for government clients. The company's focus on defense contracts and work with agencies like ICE has been a significant part of its recent business. AI

Meitu's AI research arm, MT Lab, has had six papers accepted into major international conferences including ICLR, CVPR, and ICML. One paper on scene text editing, accepted by ICML 2026, has already been integrated into Meitu Design Room and Meitu Xiuxiu PC as a 'seamless text modification' feature. This new functionality supports multiple languages and maintains visual consistency without obvious editing marks. AI

Hanwang Technology has launched the M6, a device that combines recording, note-taking, and reading functionalities. The M6 supports real-time translation for 51 languages, enabling seamless cross-lingual meeting experiences. It integrates Hanwang's proprietary 'Tiandi' large model, along with other models like DeepSeek and Tongyi Qianwen, to provide AI assistance for tasks such as summarizing meeting highlights and drafting documents. AI

Researchers have introduced a new framework for open-world object counting, addressing the brittleness of current vision-language models in accurately identifying and counting objects based on user intent. They propose redefining counting as a multi-grained problem, where both visual examples and detailed text prompts, including negative prompts, specify the target appearance and semantic granularity. To overcome the data limitations for this approach, they developed an automated pipeline using 3D synthesis and VLM filtering to create KubriCount, the largest dataset for counting tasks. Their new model, HieraCount, leverages both text and visual exemplars to significantly improve multi-grained counting accuracy and generalize to real-world scenarios. AI

This article explores the concept of Recursive Self-Improvement (RSI) by proposing a novel metric, the RSI for RSI. The author details the development and application of this metric, aiming to provide a quantitative measure for assessing the effectiveness of self-improving AI systems. The work contributes to the theoretical understanding of AI advancement and its potential for accelerated progress. AI

DBSCAN is a clustering algorithm that identifies dense regions of data points to discover arbitrary shapes. It groups together points that are closely packed, marking outliers as noise. This method is particularly effective for finding clusters of varying densities and complex structures within datasets. AI

Researchers have introduced Key-Value Means (KVM), a new attention mechanism for transformers that can handle both fixed-size and growing states. When implemented with a fixed-size cache, KVM functions as an O(N) chunked RNN with minimal parameter additions. A growable KVM cache version demonstrates competitive performance on long-context tasks, offering subquadratic prefill time and sublinear state growth. This approach is compatible with standard operations, supports chunk-wise parallelizable training, and provides a flexible trade-off between prefill time complexity and memory usage. AI

A total solar eclipse is set to occur on August 12, 2026, traversing parts of Greenland, Iceland, and Spain. Unlike the recent North American eclipse, this event will feature a low sun angle, potentially creating dramatic AI

A developer fine-tuned Google's Gemma 4 E4B model into a bias judge for approximately $30, a process that took two weeks with most of the effort focused on data pipeline construction rather than GPU time. The resulting model, capable of running locally in 30 seconds, evaluates pairs of responses to identify social bias using the Bias Benchmark for QA (BBQ) dataset. The developer encountered challenges with classification leaks, data ceilings imposed by the BBQ dataset, and disagreements among different LLMs used for labeling, ultimately leading to a refined data construction strategy. AI

Shivnath Tathe has developed "opensmith," a local-first tool designed to trace and debug LLM pipelines without sending data to the cloud. This alternative to services like LangSmith allows developers to monitor function calls, latency, token usage, costs, and errors directly on their machine. The tool gained significant traction, with over 600 downloads in its first day, indicating a strong developer demand for privacy-focused, offline observability solutions in LLM application development. AI