Google is implementing SynthID watermarks for photos modified with its Magic Editor AI feature. The system, developed by Google’s DeepMind team, embeds digital metadata tags to identify AI-altered images. According to Jess Weatherbed writing for The Verge, the feature will roll out this week. While the watermark is invisible to the naked eye, it can …
Security researchers at NowSecure have identified multiple severe security and privacy vulnerabilities in the DeepSeek iOS mobile application, which has been among the top downloaded apps since January 2025. The investigation revealed that the app transmits unencrypted user data, employs weak encryption methods with hardcoded keys, and sends sensitive information to servers connected to Chinese …
ByteDance has developed an AI system that can generate realistic full-body videos from a single photograph. The new technology, called OmniHuman, can create videos of people speaking, singing, and moving naturally with matching gestures. As reported by Michael Nuñez, the system was trained on over 18,700 hours of human video data. OmniHuman combines multiple input …
French AI company Mistral has released mobile versions of its AI assistant Le Chat for iOS and Android devices. As reported by Romain Dillet for TechCrunch, the app offers a conversation-like interface for interacting with Mistral’s language models. The company also introduced a Pro subscription at $14.99 per month, providing access to their most advanced …
Google has begun internal testing of a new AI Mode for its search engine that aims to provide more comprehensive answers to complex queries. As reported by Abner Li for 9to5Google, the feature uses Gemini 2.0 to handle exploratory questions and follow-up queries that traditional search results don’t address well. The new interface appears alongside …
Researchers from Stanford and the University of Washington have developed an AI model called s1 that rivals the capabilities of expensive commercial AI systems while costing less than $50 in computing resources to train. The model, which was created through a process called distillation using Google’s Gemini 2.0 Flash Thinking Experimental model, demonstrates similar performance …
Microsoft is extending OpenAI’s o1 reasoning model to all Copilot users at no additional cost, Tom Warren reports. The feature, called “Think Deeper,” enables more complex question handling and provides detailed step-by-step answers. It was previously available only to premium subscribers. The tool takes approximately 30 seconds to analyze questions from multiple perspectives. Microsoft AI …
The Allen Institute for Artificial Intelligence (Ai2) has released Tulu 3 405B, a new AI language model that, according to the institute’s internal testing, outperforms several leading AI systems including DeepSeek V3 and matches capabilities with OpenAI’s GPT-4o on certain benchmarks. The model contains 405 billion parameters and required 256 GPUs running in parallel for …
French startup Mistral AI has announced the release of Mistral Small 3, a 24-billion-parameter language model that the company claims matches the performance of models three times its size. According to Mistral AI, the new model achieves 81% accuracy on standard benchmarks while processing 150 tokens per second, making it comparable to Meta’s Llama 3.3 …
Alibaba’s Qwen team has launched two significant AI models – Qwen2.5-VL and Qwen2.5-Max – that demonstrate advanced capabilities in various tasks. According to the company, these models can perform text and image analysis, control computers and mobile devices, and compete with established AI systems from OpenAI, Anthropic, and Google on multiple benchmarks. The Qwen2.5-VL model …