Researchers have announced Gemma Scope, a comprehensive, open suite of sparse autoencoders for language model interpretability. This tool helps the safety community shed light on the inner workings of language models, which are often a mystery even to the researchers who train them. Gemma Scope uses sparse autoencoders as a 'microscope' to decipher the inner workings of language models.
AIGoogle | Rating: 71 | 2024-07-31 03:59:30 PM |
Google DeepMind will present over 80 research papers at the 2024 International Conference on Machine Learning (ICML) in Vienna, Austria, from July 21-27. The conference will feature the latest advances in artificial intelligence (AI) research, including multimodal generative AI, and showcase products such as Gemini Nano, LearnLM, and TacticAI.
AIGoogle | Rating: 73 | 2024-07-19 09:36:11 AM |
Project Astra is a universal AI agent that can process multimodal information, understand context, and respond naturally in conversation. It builds on the Gemini models and is designed to be truly useful in everyday life. The demo shows the prototype running on a Google Pixel phone and a prototype glasses device, with all demos taken in one continuous take in real-time.
2024-07-09 02:26:24 PM |
SynthID, a toolkit, embeds digital watermarks into AI-generated images, audio, text, or video, making them imperceptible to humans but detectable for identification. This technology is critical to promoting trust in information and is currently launched in beta, evolving and being integrated into various products.
2024-06-26 09:05:51 AM |
Veo is a generative video model that generates high-quality, 1080p resolution videos that can go beyond a minute. It provides an unprecedented level of creative control, understanding prompts for various cinematic effects. The model will help create tools that make video production accessible to everyone, suitable for filmmakers, creators, and educators.
2024-06-26 09:05:50 AM |
Imagen 3 is a text-to-image model that generates images with better detail, richer lighting, and fewer artifacts. It has improved understanding of prompts, allowing it to generate various visual styles and capture small details. Imagen 3 will be available in multiple versions, optimized for different tasks, and is initially available as a private preview for select creators in ImageFX, with a waitlist option. It will also be coming soon to Vertex AI.
2024-06-26 09:05:50 AM |
Researchers have developed a video-to-audio technology that combines video pixels with natural language text prompts to generate rich soundtracks for silent videos. This technology can be paired with video generation models to create synchronized audiovisual content.
2024-06-17 04:05:04 PM |
The UK Government hosted the first global Summit on frontier AI safety at Bletchley Park, focusing on rapid AI development and delivering concrete international action to respond to potential future risks, including the Bletchley Declaration and new AI Safety Institutes.
2024-05-20 06:15:08 AM |
Google DeepMind has introduced a safety framework to analyze and mitigate risks posed by advanced AI models. The framework aims to ensure responsible AI development and deployment, addressing potential risks and challenges as AI advances.
2024-05-17 02:05:08 PM |
Synt introduces SynthID, a watermarking method for AI-generated text and video. The technology aims to identify and track AI-generated content, ensuring authenticity and preventing misuse. The company plans to integrate SynthID into key Google products and generative AI tools.
2024-05-14 06:05:15 PM |