AI News — Friday, March 27, 2026
Google's Gemini now enables users to transfer their chat histories and personal data directly from other chatbot platforms, aiming to simplify migration and enhance user experience.
Wikipedia has implemented stricter policies to curb the use of AI in writing and editing articles, emphasizing human oversight and factual accuracy for its content.
Salesforce has introduced an enhanced Slackbot AI agent, intensifying its competition with Microsoft and Google in the rapidly growing market for AI-powered workplace tools.
Google Translate now offers a feature on iOS that turns headphones into real-time personal translators, enhancing communication across language barriers.
CUA-Suite introduces a massive, human-annotated video demonstration dataset designed to train AI agents for complex computer-use tasks, pushing the boundaries of autonomous interaction.
OpenAI has reportedly abandoned its experimental 'erotic mode' for ChatGPT, signaling a continued focus on responsible AI development and content moderation.
Google's AI Blog highlights Gemini 3.1 Flash Live, a new development aimed at making audio AI interactions more natural and reliable.
OpenAI has published insights into its strategy for developing a 'Model Spec,' outlining its commitment to defining and aligning AI behavior for safety and ethical use.
Researchers present DA-Flow, a novel method utilizing diffusion models for more accurate optical flow estimation, particularly in the presence of image degradation.
SIMART proposes a method to automatically decompose complex 3D meshes into articulated, simulation-ready assets using Multimodal Large Language Models (MLLMs), streamlining virtual environment creation.
A new perspective suggests that the evolution of coding will increasingly prioritize communication skills over mere code generation, especially with the rise of AI assistants.
An article explores a practical, engineering-focused approach to managing AI agent memory, moving beyond theoretical discussions to implement robust, scalable solutions.
This discussion delves into the trade-offs between speed and intelligence in AI coding agents, examining which factor is more crucial for effective software development.
EVA introduces an efficient reinforcement learning framework designed for end-to-end video agents, enabling them to learn and perform tasks directly from video inputs.
T-MAP presents a novel red-teaming framework that uses trajectory-aware evolutionary search to identify vulnerabilities and improve the robustness of LLM agents.