AI News — Thursday, September 17, 2026
Researchers introduce a novel approach where continual learning mechanisms are composed to achieve long-horizon memorization, addressing a critical challenge in AI systems.
Google AI announces new capabilities with Gemini 3.8 Live and 3.5 Transcribe, enabling developers to create real-time voice applications with enhanced accuracy and speed.
Industry leaders at TechCrunch Disrupt 2026 discuss the growing trend of AI agents becoming integral team members in startups and the implications for business operations.
This paper explores the transformative potential of foundation models in revolutionizing AI applications within the gaming industry, from NPC behavior to content generation.
A technical report details StepAudio 3, showcasing advancements in real-time audio generation that could significantly impact creative and interactive applications.
This paper delves into the theoretical and practical challenges of building AI capable of genuine recursive self-improvement, potentially leading to advanced general intelligence.
ScienceIDE proposes transforming the global scientific codebase into learnable environments for AI agents, aiming to accelerate scientific discovery and research.
The openarm project releases a fully open-source humanoid arm designed for physical AI research and deployment, particularly in environments requiring complex physical interaction.
Al Gore offers a nuanced perspective on AI risks, suggesting that the primary concerns lie beyond the environmental impact of data centers, hinting at broader societal implications.
A companion technical report for StepAudio 3 focuses specifically on its capabilities for music generation, detailing new methods and results.
Researchers propose Grouped Value Attention, an efficient KV caching mechanism that uses on-demand key reconstruction to improve performance in large language models.
A new formal framework, Generalized Agent Iteration, is introduced to unify iterative policy improvement and recursive self-improvement, advancing the theoretical understanding of AI agents.
LynnReal-Omni presents a new approach to native multi-modal video generation, specifically designed to support agentic visual workflows with high fidelity.
Google AI outlines its initiatives and research focus on leveraging artificial intelligence to address pressing societal challenges and create positive global impact.
OpenAI details its efforts and strategies to make AI more accessible and beneficial for older adults, integrating AI tools into their daily routines.