AI News — Saturday, April 18, 2026

Sam Altman's Project World is reportedly scaling its human verification services, with Tinder being the latest platform to integrate its technology.

Sources indicate that Cursor is in talks to raise over $2 billion at a staggering $50 billion valuation, driven by rapid growth in its enterprise AI solutions.

Kevin Weil and Bill Peebles have exited OpenAI, signaling the company's continued effort to streamline operations and shed non-core projects.

OpenAI has announced a new initiative focused on accelerating the cyber defense ecosystem, aiming to enhance protection against digital threats.

Researchers introduce HY-World 2.0, an advanced multi-modal world model capable of reconstructing, generating, and simulating complex 3D environments.

Google is rolling out seven new AI-powered features to help users plan and experience smarter summer travel, leveraging advancements in Google Search and AI.

The Gemini app now offers enhanced ways for users to create personalized images, further expanding its creative AI functionalities.

RAD-2 presents a novel generator-discriminator framework designed to significantly scale reinforcement learning, offering new avenues for complex AI training.

GlobalSplat introduces an efficient feed-forward method for 3D Gaussian Splatting, utilizing global scene tokens to enhance performance in 3D rendering and reconstruction.

NuHF Claw introduces a risk-constrained cognitive agent framework designed to provide human-centered procedure support in digital nuclear control rooms, enhancing safety and efficiency.

RadAgent is presented as a tool-using AI agent capable of providing stepwise interpretations of chest computed tomography scans, potentially aiding medical diagnostics.

OpenAI has published an academy guide detailing how users can effectively create images using ChatGPT, making the feature more accessible.

A new resource from OpenAI Academy provides insights and best practices for leveraging ChatGPT as a tool in academic and professional research.

This paper investigates the mechanisms of memory transfer learning, exploring how memories are effectively transferred across different domains in AI coding agents.

A new framework proposes simulating human cognition through heartbeat-driven autonomous thinking activity scheduling for large language model-based AI systems.