AI News — Thursday, May 14, 2026

Notion has integrated AI agents directly into its workspace, allowing users to automate tasks and enhance productivity within their existing workflows.

xAI is reportedly operating a large number of gas turbines at its Mississippi data center without proper regulatory oversight, raising environmental and safety concerns.

Salesforce has launched a new Slackbot AI agent, intensifying its competition with Microsoft and Google in the rapidly evolving workplace AI market.

A key figure at Anthropic envisions a future where AI systems will proactively understand and address user needs even before they are consciously recognized.

This paper explores 'World Action Models' as a critical advancement for embodied AI, enabling agents to better understand and interact with complex environments.

The workflow of Claude Code's creator has been unveiled, captivating developers with its innovative approach to AI-assisted coding.

AlphaGRPO introduces a novel method for self-reflective multimodal generation in Universal Multimodal Models by using a decompositional verifiable reward system.

OpenAI details the development of a secure and functional sandbox environment to facilitate the integration and use of Codex on Windows operating systems.

Google has enhanced the Gemini API with webhooks, allowing developers to reduce friction and latency for long-running asynchronous tasks.

This article discusses strategies for optimizing and reducing the size of 'Bloated MCP' (likely a software project or model) using a 'Code Mode' approach.

ToolCUA proposes a framework for optimal orchestration of GUI-tool paths, aiming to improve the efficiency and effectiveness of computer use agents.

AWS Lambda now supports file systems, enabling developers to deploy and run AI agents directly within serverless functions for more complex operations.

L2P presents a new technique designed to unlock the latent potential within models for more effective and high-quality pixel generation.

This research introduces Multi-Stream LLMs, a novel architecture that uses parallel streams for thoughts, inputs, and outputs to enhance language model efficiency and capability.

OpenAI shares insights gained from 'Parameter Golf,' a project that explored the effectiveness and challenges of AI-assisted research methodologies.