The Open ASR Leaderboard now includes multilingual and long-form transcription tracks, revealing that Conformer encoders with LLM decoders achieve the best accuracy, while closed-source models still lead in long-form tasks. Speed–accuracy tradeoffs and performance drops in multilingual settings highlight key challenges in modern speech recognition.

Maggie Hulce, Indeed’s CRO, discusses how AI is reshaping job search, recruiting, and talent acquisition by improving efficiency and personalization for both employers and job seekers.

Google has released Gemini 3.1 Flash-Lite, a cost-effective AI model designed for high-scale intelligence and efficient performance in large workloads. It represents an advancement in balancing affordability and capability for widespread AI deployment.

AI-equipped drones are being used by scientists to study the endangered Māui dolphins in New Zealand, combining drone surveillance, artificial intelligence, and cloud technology to gather critical conservation data. This approach is part of a broader trend using AI and digital tools to enhance wildlife monitoring and protect imperiled species globally.

The article introduces M-JudgeBench, a comprehensive, capability-oriented benchmark for evaluating multimodal large language models (MLLMs) as judges, and proposes Judge-MCTS to generate high-quality, diverse training data through MCTS-driven methods. It addresses limitations in current benchmarks by focusing on fine-grained judgment capabilities like reasoning consistency, length bias avoidance, and error detection.

This paper proposes pause-aware decoding strategies using multimodal large language models to generate real-time, well-timed game video commentary without fine-tuning. The dynamic interval-based approach outperforms fixed-interval methods by aligning generated commentary more closely with human timing and content in racing and fighting games.

The article promotes GitHub Copilot Dev Days events, highlighting resources and opportunities for developers to learn about AI and machine learning tools, including generative AI and LLMs, and how they can improve coding workflows.

Vertex AI offers a comprehensive platform for generative AI, featuring access to Google’s latest models like Gemini 3 and Imagen 4, along with tools for prompt engineering, model deployment, and integration via APIs and SDKs. It supports seamless migration from previous platforms, provides detailed documentation, and enables developers to build and deploy AI applications across cloud environments.

Project Genie enables users to create interactive worlds using text and images, and this article offers four tips for writing effective prompts to generate compelling experiences. The tool leverages AI to transform creative inputs into dynamic, explorable environments.