Google DeepMind has introduced DiffusionGemma, a new AI model that generates text up to 4x faster than traditional models. This model uses a diffusion-based approach, processing entire blocks of text simultaneously, and can achieve up to 1000 tokens per second on an NVIDIA H100.
DiffusionGemma is designed for speed-critical local workflows and can be fine-tuned for specific tasks. It runs efficiently on various NVIDIA hardware, including H100, DGX Spark, and RTX PRO, and can be accessed via Hugging Face, NVIDIA NIM, and NVIDIA NeMo AutoModel.
In other AI news, OLAY and Haut.AI have collaborated to introduce a new Virtual Companion Technology to the OLAY Skin Advisor experience, using generative AI and clinical data modeling to simulate skincare routine performance.
Meanwhile, Germany is setting up an AI Safety Institute to analyze the capabilities and risks of modern AI models, focusing on their impact on cybersecurity. VivaTech 2026 will focus on enterprise AI, discussing deployment challenges, governance, and security.
Google's AI assistant Gemini recently experienced service issues, but mitigations have been applied to reduce the impact. There are also concerns about AI in Black communities, including fears of exploitation and resource diversion.
Key Takeaways
- Google DeepMind releases DiffusionGemma, a 26B MoE AI model that generates text up to 4x faster than traditional models.
- DiffusionGemma achieves up to 1000 tokens per second on NVIDIA H100 and can run on local hardware like Nvidia DGX or gaming GPUs.
- OLAY and Haut.AI collaborate on Virtual Companion Technology, using generative AI and clinical data modeling for skincare routines.
- Germany sets up an AI Safety Institute to analyze AI capabilities and risks, focusing on cybersecurity.
- VivaTech 2026 to focus on enterprise AI, discussing deployment challenges, governance, and security.
- Google's Gemini AI service experiences issues, but mitigations have been applied.
- Concerns about AI in Black communities include fears of exploitation and resource diversion.
- Forrester's 2026 Threat Intelligence Report highlights top risks, including AI agents and software supply chain risks.
- DiffusionGemma can be accessed via Hugging Face, NVIDIA NIM, and NVIDIA NeMo AutoModel.
- Panache Digital Games apologizes for using AI-generated assets in their game, 1666: Amsterdam.
Google's DiffusionGemma AI model is 4x faster
Google DeepMind has released a new AI model called DiffusionGemma, which generates text 4x faster than traditional models. It uses a diffusion-based approach, processing entire blocks of text simultaneously, and can achieve up to 1000 tokens per second on an NVIDIA H100. The model is designed for speed-critical local workflows and can be fine-tuned for specific tasks.
Run DiffusionGemma on NVIDIA for high-speed text generation
DiffusionGemma, developed by Google DeepMind and optimized for NVIDIA platforms, generates text tokens in parallel, achieving high throughput and lower serving costs. The model runs efficiently on various NVIDIA hardware, including H100, DGX Spark, and RTX PRO, and can be accessed via Hugging Face, NVIDIA NIM, and NVIDIA NeMo AutoModel.
Google's DiffusionGemma speeds up text generation
Google has introduced DiffusionGemma, an experimental open model that explores text diffusion for faster text generation. The 26B Mixture of Experts (MoE) model generates entire blocks of text simultaneously, delivering up to 4x faster text generation on GPUs. It prioritizes speed but has lower output quality than standard Gemma 4.
Google releases DiffusionGemma for faster AI
Google DeepMind has released DiffusionGemma, a new AI model that generates text 4x faster than traditional models. It uses diffusion-based denoising to produce entire blocks of text in parallel and can run on local hardware like Nvidia DGX or gaming GPUs.
Google AI releases DiffusionGemma for faster text generation
Google AI has released DiffusionGemma, a 26B MoE open model that uses text diffusion for up to 4x faster generation. The model generates entire blocks of text simultaneously and can achieve up to 1000 tokens per second on a single NVIDIA H100.
Codex creates campaign assets for Maison Fève
Codex AI generates campaign concepts and visual assets for brands like Maison Fève, streamlining marketing production. The video demonstrates how AI can be instrumental in creating campaign concepts and assets, transforming initial ideas into tangible marketing materials.
Developers apologize for using AI in game
Panache Digital Games apologized for using AI-generated assets in their game, 1666: Amsterdam. The studio admitted to using AI but promised that the final game will not include AI-generated assets.
OLAY and Haut.AI collaborate on virtual companion technology
OLAY and Haut.AI have collaborated to introduce a new Virtual Companion Technology to the OLAY Skin Advisor experience. The technology uses generative AI and clinical data modeling to simulate how a recommended skincare routine is expected to perform over time.
The dark side of AI obsession
The article discusses the potential dark side of AI obsession, highlighting the environmental impact and potential biases of AI models. It also mentions the importance of considering the social and environmental implications of AI development.
Germany to set up AI Safety Institute
Germany's National Security Council has decided to set up an AI Safety Institute to analyze the capabilities and risks of modern AI models. The institute will assess how advanced AI models affect cybersecurity in Germany.
Enterprise AI focus at VivaTech 2026
VivaTech 2026 will focus on enterprise AI, discussing the challenges of deploying AI in large organizations, governance, compliance, security, and operational reliability. The event will bring together founders, investors, and enterprise leaders to discuss AI's transition from experimentation to production.
Gemini AI service issues
Google's AI assistant Gemini experienced service issues, with many users encountering errors. Google applied mitigations to reduce the impact, and the majority of users should no longer observe issues.
AI concerns in Black communities
The article discusses concerns about AI in Black communities, including fears that AI may be exploitative and take resources away from already burdened communities.
Forrester 2026 Threat Intelligence Report
Forrester's 2026 Threat Intelligence Report highlights top risks, including AI agents, nation-state threats, and AI software supply chain risks. The report emphasizes the need for CISOs to address these threats and develop strategies to mitigate them.
Sources
- DiffusionGemma: Google's AI is 4x Faster
- Run DiffusionGemma on NVIDIA for Developer-Ready, High-Throughput Text Generation
- DiffusionGemma: 4x faster text generation
- Google DeepMind releases DiffusionGemma, a model that runs local AI 4x faster
- Google AI Releases DiffusionGemma, a 26B MoE Open Model Using Text Diffusion for Up to 4x Faster Generation
- Codex Creates Campaign Assets for Maison Fève
- Once again players are right to suspect AI was used in a game, once again a dev apologizes for using AI in their game
- Haut.AI Collaborates with OLAY on Virtual Companion Technology
- Why Our Obsession With Artificial Intelligence Might Be Completely Wrong - Futura-Sciences
- Germany's National Security Council greenights an AI Safety Institute modeled after the UK's AISI
- Why enterprise AI will be a major focus at VivaTech 2026
- Gemini Is Down? Live Updates on Google Workspace's AI Errors
- Are Black People Being Left Behind By AI?
- Forrester 2026 Threat Intelligence Report: AI Agents Top CISO Risk List
Comments
Please log in to post a comment.