Artificial Intelligence: News, Business, Science

AI research

Aug 5, 2025Aug 5, 2025

Google Deepmind's Genie 3 creates interactive 3D worlds that stay consistent for "multiple minutes"

AI in practice

Aug 1, 2025Aug 1, 2025

Google upgrades Gemini with Deep Think and flags early warning risks

AI research

Jul 30, 2025Jul 30, 2025

OpenAI’s math breakthrough might also mean AI is getting better at knowing its own limits

AI research AI and society AI in practice

AI in practice

Aug 7, 2025Aug 7, 2025

An invisible prompt in a Google Doc made ChatGPT access data from a victim’s Google Drive

AI in practice

Aug 7, 2025Aug 7, 2025

Short

Google is rolling out a new "Guided Learning" feature in the Gemini app, designed to help users break down complex topics step by step. Unlike standard AI answers, Guided Learning uses follow-up questions, interactive explanations, images, videos, and quizzes to deepen understanding. Similar learning tools are already available from OpenAI and Anthropic. The feature is powered by LearnLM, a model optimized specifically for education. According to Google, Guided Learning was developed with input from educators, learning scientists, and students. Teachers can also add Guided Learning links directly into Google Classroom.

Video: Google

Google

AI in practice

Aug 7, 2025Aug 7, 2025

A deleted GitHub post gives an early look at OpenAI’s next major model, GPT-5

THE DECODER

Aug 6, 2025Aug 6, 2025

Short

Google search chief Liz Reid says AI-generated answers are not lowering website traffic. In a blog post, she writes that organic clicks have stayed "relatively stable" over the past year and that Google is now sending "slightly more quality clicks"—meaning users stay on a site instead of quickly returning. Reid argues that AI helps improve questions and results.

"The web has existed for over three decades, and we believe we’re entering its most exciting era yet."

Liz Reid

Her claims conflict with outside reports showing major traffic drops tied to AI answers in search results. Reid rejects those findings, saying they rely on flawed methods, older data, or single examples not tied to AI features. Google has not released detailed traffic or click-through numbers for its AI features.

Google

AI in practice

Aug 6, 2025Aug 6, 2025

Short

Google has officially launched Jules, its AI coding agent that was in testing since May. Jules can handle multiple tasks at once and automate routine developer work, similar to Anthropic's Claude Code. The latest version includes easier controls, lets users reuse previous tasks, and adds visual feedback for web app testing, Google says. Jules runs on Gemini 2.5 and is available through several Gemini subscription tiers.

Google is also rolling out Gemini CLI GitHub Actions, a free AI tool that integrates directly with GitHub repositories, automating routine developer tasks on demand. The agent responds to events like new issues or pull requests, working in the background based on the context of each project. Google says the agent supports authentication without fixed credentials and logs activity via OpenTelemetry. Technically, it builds on Gemini CLI.

Jules Github CLI Google Blog

AI in practice

Aug 6, 2025Aug 6, 2025

Short

Anthropic has released a new open source tool on GitHub that automatically checks code for security vulnerabilities. The GitHub action "Claude Code Security Reviewer" uses the Claude AI model to scan pull requests for potential security issues.

According to the project description, the tool can spot security vulnerabilities across different programming languages by understanding the context of the code. It automatically adds comments directly in code discussions, filters out likely false positives, and focuses only on files that have been modified. The tool is available under the MIT license on GitHub.

Github

AI research

Aug 6, 2025Aug 6, 2025

Alibaba's new Qwen-Image model generates high-fidelity text inside images

AI in practice

Aug 6, 2025Aug 6, 2025

Short

Microsoft has unveiled Project Ire, an AI system designed to automatically analyze software files and determine whether they contain malware. The system uses reverse engineering tools to inspect program code, builds a chain of evidence, and then decides if a file is dangerous. Project Ire was developed by teams from Microsoft Research, Defender Research, and Discovery & Quantum. In tests, Ire correctly identified 90 percent of malware in Windows drivers while producing very few false positives. In another trial with files that were especially difficult to classify, the system achieved 89 percent precision. Microsoft plans to integrate Project Ire into Microsoft Defender as a "Binary Analyzer," aiming to help security analysts by automating routine threat detection.

Microsoft