DeepMind Launches Sign Language-to-Text AI Model for Accessibility
DeepMind introduces SL2T, a new AI model that converts sign language to text for Deaf and hard of hearing users.
Sources · DeepMind
The news desk
Follow what’s changing.
Recent developments in artificial intelligence, with the original sources one click away.
All news
381–400 of 1,897 Organizations Face ROI Challenges in AI Agent Adoption Despite Rapid Implementation
Companies are quickly adopting AI agents but struggle to achieve desired ROI, according to MIT Technology Review analysis.
Sources · MIT Technology Review
Research Reveals Large Language Models Easily Abandon Correct Beliefs Under Persuasion
New study shows single persuasive arguments can collapse LLM accuracy to near zero, even when arguments are factually false.
Sources · arxiv.org
Researchers Advance AI Reasoning Capabilities Through Training-Free and Agentic Approaches
New research demonstrates multi-agent systems and agentic reasoning frameworks that improve AI performance on specialized tasks without traditional training.
Sources · arxiv.org
White House A.I. Plan Details Leaked to Media Amid Official Silence
Details from the White House's artificial intelligence plan have leaked to news outlets, though the administration has shared little officially.
Sources · NYT Technology
New Benchmarks Target LLM Reasoning Capabilities Across Multiple Domains
Researchers released four new benchmarks evaluating LLM performance in computer architecture, moral reasoning, chain-of-thought monitoring, and scientific visualization.
Sources · arxiv.org
OpenAI COO Brad Lightcap Steps Down to Start New Venture
Brad Lightcap, OpenAI's longtime chief operating officer, is leaving the company to pursue a new venture, marking another executive departure.
Sources · NYT Technology / TechCrunch AI
OpenAI Launches ChatGPT Desktop App for Linux
OpenAI has released a dedicated ChatGPT desktop application for Linux operating systems.
Sources · TechCrunch AI
OpenAI's Daybreak Cybersecurity Models Now Available on Amazon Bedrock
OpenAI has made its Daybreak Red and Blue cybersecurity models available through AWS Amazon Bedrock for eligible customers.
Sources · openai.com / aws.amazon.com
Researchers Develop Safety-Focused AI Systems for Healthcare, Public Services, and Interior Design
New arXiv papers present specialized AI architectures prioritizing safety, efficiency, and structured reasoning over general-purpose LLM approaches.
Sources · arxiv.org
Model ML Leverages GPT-5.6 Sol for Streamlined Finance Workflows
Model ML utilizes OpenAI's GPT-5.6 Sol to automate finance tasks from research to creating editable PowerPoint and Excel deliverables.
Sources · OpenAI
Platforms Begin Implementing Tools to Combat AI-Generated Content
Growing number of platforms introduce policies to flag, label, and ban AI-generated content in response to user backlash.
Sources · WIRED AI
Researchers Propose Graph-Based Framework for Medication Safety in Older Adults with Multimorbidity
New ATLAS framework uses coupled graph-policy distillation to help LLM agents make safer medication recommendations for elderly patients.
Sources · arxiv.org
Researchers Release Four New Benchmarks for Evaluating LLM Reasoning Capabilities
New benchmarks test LLM performance in computer architecture, moral reasoning, chain-of-thought monitoring, and scientific visualization tasks.
Sources · arxiv.org
Startups Pursue Next Generation of Large Language Models Beyond Transformers
Companies are exploring alternatives to transformer architecture that powers current AI models, seeking improvements in efficiency and capabilities.
Sources · MIT Technology Review
Anthropic Enables Auto Mode by Default in Claude Code
Anthropic is making Claude Code's auto mode the default setting, reducing the need for human oversight during programming tasks.
Sources · TechCrunch AI
Capek 0.5: Execution-Centric Vision-Language Model Organizes Embodied AI Capabilities by Function
New embodied AI model organizes vision-language capabilities around execution roles rather than isolated tasks, using specialist training and merging.
Sources · arxiv.org
Meetily Offers Free, Open-Source Meeting Transcription Alternative
New tool provides AI-powered meeting transcription and summarization without requiring a paid subscription.
Sources · WIRED AI
New Research Proposes MARL-Centered Architecture for LLM Integration in Manufacturing
Researchers present a reference architecture combining multi-agent reinforcement learning with large language models for smart manufacturing systems.
Sources · arxiv.org
Three New Research Papers Advance Long-Horizon AI Agent Systems
Researchers release benchmark for agent trajectory attribution, recursive task synthesis framework, and tree search method for complex systems design.
Sources · arxiv.org
