NO/FOMO

Independent AI signal, once a day

The AI briefing worth opening.

ISSUE DATE2026-05-03ENGLISH EDITION
This issue
—
All time
—

Hacker News

6 stories
01

OpenAI's o1 correctly diagnosed 67% of ER patients vs. 50-55% by triage doctors

A recent report indicates that OpenAI's AI system, designated as 'o1', has demonstrated superior diagnostic capabilities in emergency room (ER) patient triage compared to human doctors. In a trial, the AI achieved a correct diagnosis rate of 67%, significantly outperforming the 50-55% accuracy range observed among human medical professionals responsible for initial patient assessments. This finding, based on an alleged Harvard trial as referenced in a Guardian article, underscores the potential for advanced artificial intelligence to revolutionize diagnostic processes within critical healthcare environments. The deployment of AI systems like 'o1' could lead to enhanced operational efficiency, reduced human error in triage, and ultimately improved patient outcomes by facilitating faster and more accurate preliminary diagnoses in high-pressure medical situations. The implications for the integration of AI into clinical practice are profound, suggesting a future where AI supports and augments critical medical decision-making.

AI diagnosismedical AIemergency medicinetriageOpenAIArtificial IntelligenceMachine LearningAI Agent
02

Talking to Transformers

The article 'Talking to Transformers' explores the intricate mechanisms and practical methodologies for effectively engaging with and understanding Transformer-based artificial intelligence models. It likely delves into various interaction paradigms, including advanced prompt engineering techniques, methods for interpreting complex generative outputs, and strategies for leveraging the models' inherent capabilities for sophisticated dialogue and task execution. The discussion would meticulously examine the architectural nuances of Transformers, such as their self-attention mechanisms, which are fundamental to their advanced language comprehension and generation prowess. Furthermore, it may address the evolving landscape of human-AI collaboration, presenting best practices for optimizing communication with these powerful neural networks. The piece likely provides insights into mitigating common challenges, enhancing model responsiveness, and ensuring ethical deployment, ultimately aiming to equip readers with a comprehensive understanding of effective strategies for interactive AI systems built upon the Transformer architecture, crucial for advancements in natural language processing and large language models.

TransformersNatural Language ProcessingLarge Language ModelsPrompt EngineeringAI InteractionArtificial IntelligenceLarge Language ModelNatural Language Processing
03

Show HN: Apple's SHARP running in the browser via ONNX runtime web

A developer has successfully implemented Apple's SHARP, a single-image 3D Gaussian splatting model, to run directly in a web browser using ONNX runtime web with WebGPU as the execution provider. This project demonstrates the feasibility of executing complex machine learning models client-side, eliminating the need for server-side processing. Users can upload an image, and the browser-based application generates a .ply file for download or live preview, ensuring privacy as images never leave the user's tab. While the initial load can be slow due to the model's substantial size (around 2.4 GB), subsequent inference typically completes in a few seconds on modern hardware. The project utilizes weights released by Apple under a research-use-only license, with the exported ONNX model hosted on R2 for convenience.

SHARP3D Gaussian SplattingONNX Runtime WebWebGPUIn-browser MLArtificial IntelligenceMachine LearningComputer Vision
04

Underwater robot tracks sperm whale conversations in real time

A groundbreaking development in marine technology has seen an advanced underwater robotic system successfully deployed to track and analyze sperm whale conversations in real-time. This innovative system combines state-of-the-art autonomous underwater vehicle (AUV) capabilities with highly sensitive acoustic sensors and sophisticated bioacoustics processing algorithms, likely employing artificial intelligence for pattern recognition. The ability to monitor whale communication instantaneously offers an unprecedented window into the intricate social structures, foraging behaviors, and migratory patterns of these elusive marine mammals. This real-time data acquisition is expected to significantly enhance marine biology research, providing critical insights for conservation strategies and understanding the impact of environmental changes and human activities on cetacean populations. The project represents a substantial leap in leveraging robotics and AI for ecological monitoring, overcoming the inherent challenges of underwater data collection and potentially paving the way for decoding complex animal communication, thereby deepening our appreciation and efforts to protect marine biodiversity.

Underwater RoboticsBioacousticsAutonomous Underwater VehiclesReal-time MonitoringMarine ConservationRoboticsArtificial IntelligenceMachine Learning
05

How Kepler built verifiable AI for financial services with Claude

Kepler has successfully implemented verifiable artificial intelligence solutions for the financial services industry by utilizing Anthropic's Claude. This strategic move directly tackles the significant challenges of transparency, auditability, and stringent regulatory compliance inherent in the financial sector. By leveraging Claude's advanced capabilities, Kepler is able to construct AI applications that offer enhanced reliability and explainability, crucial for fostering trust and mitigating risks. Unlike conventional "black-box" AI models, the systems developed with Claude provide clear, traceable decision pathways, allowing financial institutions to understand and validate AI-driven outcomes. This verifiable AI framework ensures adherence to complex financial regulations and internal governance policies, thereby reducing operational risks and increasing stakeholder confidence. The collaboration highlights a pivotal shift towards the responsible and ethical deployment of AI within highly regulated environments, setting a new benchmark for how AI can support robust decision-making and maintain operational integrity through transparent methodologies. This initiative positions Kepler at the forefront of AI innovation in finance.

Verifiable AIFinancial ServicesLarge Language ModelClaudeAI ComplianceArtificial IntelligenceLarge Language ModelNatural Language Processing
06

ASU Using AI Tool to Create Courses from Professors' Work Without Their Knowledge

Arizona State University (ASU) is reportedly employing an artificial intelligence tool to generate new course content directly from the existing academic work of its professors. This process is allegedly occurring without the explicit knowledge or consent of the faculty members involved, raising significant concerns across the academic community regarding intellectual property and academic integrity. The deployment of such an AI tool by a major educational institution highlights the rapidly evolving landscape of technology integration in higher education, particularly concerning curriculum development and faculty relations. Critical questions are emerging regarding intellectual property rights, the ownership of educational materials, and the ethical frameworks governing the use of AI to repurpose human-created content. This initiative could set a controversial precedent for how universities manage intellectual assets and engage with their faculty in an era increasingly shaped by advanced AI capabilities, prompting a broader dialogue on academic freedom, fair compensation for original scholarly work, and the role of human educators in an AI-assisted environment. This situation underscores the urgent need for clear policies and transparent communication surrounding AI adoption in educational institutions.

AI ToolCourse CreationIntellectual PropertyHigher EducationAcademic EthicsArtificial IntelligenceGenerative AILarge Language Model