OpenAI has updated its Agents SDK, focusing on enhancing safety and capabilities for enterprises building AI agents, signaling a move towards more robust and reliable AI deployments.
Google has launched Gemini 3.1 Flash TTS, the latest generation of its text-to-speech model, promising more expressive and natural-sounding AI-generated speech.
Researchers introduce ClawGUI, a new unified framework designed to streamline the training, evaluation, and deployment of AI agents that interact with graphical user interfaces.
A new research paper presents KnowRL, a method that leverages reinforcement learning with minimal-sufficient knowledge guidance to significantly improve the reasoning capabilities of Large Language Models.
Hightouch has reached a significant milestone of $100 million in Annual Recurring Revenue, largely attributed to the success and adoption of its AI-powered marketing tools.
New research delves into the intricacies of on-policy distillation for large language models, offering insights into its mechanisms and proposing an improved recipe for more effective training.
New data from LinkedIn indicates that, contrary to some concerns, AI is not currently the primary cause of a decline in hiring, though its future impact remains a subject of observation.
A new paper introduces Nemotron 3 Super, an open and efficient mixture-of-experts model combining Mamba and Transformer architectures, designed to enhance agentic reasoning capabilities.
Researchers are exploring methods for autonomous long-horizon engineering, aiming to enable AI systems to independently conduct and manage complex machine learning research projects.
Lyra 2.0 is introduced as a new system for creating explorable generative 3D worlds, pushing the boundaries of immersive virtual environment generation.
A guide has been published detailing effective prompting techniques for Google's new Gemini 3.1 text-to-speech model, helping developers achieve high-quality expressive audio.
Audio-Omni is a new development that expands multi-modal AI understanding to include versatile audio generation and editing capabilities, bridging the gap between text, image, and sound.
A new study suggests that while AI assistance can be helpful, it may inadvertently reduce users' persistence and negatively impact their ability to perform tasks independently.
Researchers propose 'BERT-as-a-Judge' as a robust and efficient alternative to traditional lexical methods for evaluating Large Language Models based on reference texts.
A study highlights critical safety vulnerabilities in computer-use agents, revealing how seemingly benign user instructions can inadvertently expose significant security risks.