Google Successfully Attacks OpenAI GPT-3.5-turbo Model at a Cost as Low as 150 Yuan

Google's latest research has unveiled a successful method to attack OpenAI's GPT-3.5-turbo model, with costs as low as 150 RMB. The attack is both simple and effective, requiring fewer than 2000 API queries to obtain critical information. This experiment serves as a reminder that even large-scale language models can face security threats. OpenAI has already taken steps to modify the model's API to prevent future attacks.

Google's AI Programming Tool for Android Helps Users Easily Create Applications

Google has launched the Android version of AI Studio, a programming tool that simplifies app development and is now available for pre-registration on Google Play. This tool enables users to build applications using artificial intelligence by following simple instructions through an intuitive interface and smart suggestions, making it suitable for both beginners and experienced developers.

Advertising as a Service: The AI-Driven Redefinition of Google Search, Integrating Paid Recommendations into the Chat Flow

At the 2026 I/O conference, Google announced the most significant restructuring of its search business in 25 years, fully integrating Gemini 3.5 Flash and launching an AI-driven new ad format. These ads break the traditional boundaries between search ads and results, evolving from passive displays to active conversational services. With Gemini's reasoning capabilities, commercial information is deeply integrated, enabling a smarter and more natural interactive experience.

Building a Trust Defense Line: OpenAI Teams Up with Google to Deploy Dual Digital Genes to Combat AI Fake Content

OpenAI released a dual protection mechanism on May 7, 2026, to combat AI image fraud. The core includes C2PA digital traceability (metadata embedded with generation records) and invisible watermark technology, complementing each other to form a double insurance system, aiming to establish a new industry standard for verifying the authenticity of images.

Breakthrough in OpenAI Inference Model: AI Successfully Refutes Erdős Unit Distance Conjecture

On May 20, 2026, OpenAI's reasoning model overturned Paul Erdős' 1946 'unit distance conjecture,' solving a core problem in discrete geometry that had persisted for nearly 80 years. Unlike previous AI achievements, this result gained widespread academic recognition, marking a key breakthrough from AI 'retrieval' to 'original creation.'....

Google Upgrades AI Mode Search Ads: Gemini Model Deeply Involved and Launches In-Built Chatbot Ads

Google is testing a new ad format in AI mode search, using the Gemini model to generate customized interpreters for sponsored products, directly explaining the reasons to purchase, such as the compact Italian capsule coffee maker example. This move follows the launch of a larger conversational search box, marking the deep integration of Gemini capabilities into the core advertising field.