AI News — Tuesday, March 31, 2026
A new poll reveals that while more Americans are adopting AI tools, a growing number express distrust in the accuracy and reliability of the results generated by these technologies.
A recent Quinnipiac poll indicates that a significant minority of Americans, 15%, are willing to work under an AI supervisor, highlighting evolving attitudes towards AI integration in the workplace.
Researchers introduce a novel hybrid memory system designed to enhance dynamic video world models, allowing AI to better understand and predict complex, evolving visual environments over longer durations.
A new research paper unveils ShotStream, a system capable of generating multi-shot videos in a streaming fashion, paving the way for more interactive and dynamic AI-powered storytelling experiences.
AI gateway startup LiteLLM has announced its separation from Delve, a move that could impact its standing and partnerships within the rapidly evolving AI ecosystem.
OpenAI is leveraging its AI technology to assist disaster response teams across Asia, helping them translate AI insights into actionable strategies during critical situations.
A developer recounts building a local AI agent to audit their articles, discovering its surprising effectiveness in flagging issues across all their work, highlighting the potential of personalized AI assistants.
A new tool named Goose is gaining attention for providing similar code generation capabilities to Claude Code, but at no cost, potentially disrupting the market for AI coding assistants.
New research introduces PackForcing, a method demonstrating that training AI models on short video segments can effectively enable them to generate and infer from much longer video contexts, improving efficiency.
MedOpenClaw introduces auditable AI agents for medical imaging that can reason over complete, uncurated studies, promising greater transparency and reliability in clinical applications.
While AI agents are demonstrating capabilities for long-horizon tasks, experts caution that true full autopilot systems are still a distant reality, emphasizing the ongoing challenges in complex autonomous decision-making.
The developer behind Claude Code has shared their innovative workflow, sparking significant interest and discussion among the developer community eager to learn advanced AI integration techniques.
Researchers present Trace2Skill, a method that distills specific lessons from agent trajectories into broadly transferable skills, enhancing the efficiency and adaptability of AI agents in various tasks.
RealChart2Code proposes a new benchmark and approach for generating code from charts using real-world data and multi-task evaluation, significantly improving the accuracy and utility of this AI capability.
OpenClaw is introduced as a self-hosted AI assistant system, offering users greater control and privacy over their AI interactions and data.