The news desk
Recent developments in artificial intelligence, with the original sources one click away.
Four arxiv.org papers examine speculative decoding dynamics, clinical hallucination metrics, embodied agent planning, and automated reasoning structures.
Sources · arxiv.org
Google's new text-to-speech model offers granular audio tags for vocal control across 70+ languages with built-in AI audio watermarking.
Sources · blog.google / deepmind.google
Google has released a native Gemini application for Mac, enabling users to share screen content and local files for AI assistance.
Sources · TechCrunch AI
New training framework allows 4B parameter model to exceed performance of larger models on knowledge graph tasks by average of 16.7%.
Sources · arxiv.org
As AI handles more technical work, human skills like persuasion and relationship-building are becoming increasingly valuable in the workplace.
Sources · NYT Technology
Researchers introduce benchmarks and memory systems to improve LLM agents' spatial analysis, monitoring, and learning capabilities.
Sources · arxiv.org
Amazon Web Services launches a structured framework to help organizations move generative AI projects from concept to production.
Sources · Amazon AWS AI
Google adds Skills feature to Chrome, allowing users to save and reuse AI prompts across web pages with Gemini integration.
Sources · wired.com / techcrunch.com
MIT Technology Review announces its upcoming annual list highlighting technologies expected to have the biggest impact on life and work.
Sources · MIT Technology Review
Researchers introduce RT-LRM benchmark to evaluate safety risks in reasoning AI models, finding them more vulnerable than traditional language models.
Sources · arxiv.org
New hierarchical graph-based memory system aims to balance rapid context updates with stable knowledge retention in language model agents.
Sources · arxiv.org
Amazon Web Services demonstrates how AWS Lambda enables scalable reward functions for customizing Amazon Nova models using RLVR and RLAIF approaches.
Sources · Amazon AWS AI
Developers are using AI agents to simulate social interactions for optimizing choices in colleagues, friends, and romantic partners.
Sources · WIRED AI
New research tackles cross-lingual speech models, student coding behavior simulation, and brain-based language processing validation.
Sources · arxiv.org
Stanford University releases its annual AI Index report, offering data-driven insights into the current state of artificial intelligence.
Sources · MIT Technology Review
New training-free framework achieves cancer detection accuracy comparable to supervised ML while enabling smaller models to perform temporal reasoning.
Sources · arxiv.org
Research shows current leading AI models achieve only ~40% accuracy when handling conflicts among 12 levels of instructions in agent settings.
Sources · arxiv.org
Researchers propose noise-aware technique that cuts hallucination rates in auditory language models without costly fine-tuning.
Sources · arxiv.org
Four recent papers explore LLM decision-making under uncertainty, task transfer, misinformation simulation biases, and preference optimization strategies.
Sources · arxiv.org
Researchers release PinpointQA, a benchmark with 10,094 QA pairs testing multimodal AI models on finding and describing small objects in indoor videos.
Sources · arxiv.org