Google DeepMind's Genie 2 can generate interactive 3D worlds

Pivot 5: 5 stories. 5 minutes a day. 5 days a week.

1. Google DeepMind's Genie 2 can generate interactive 3D worlds

Google DeepMind has announced Genie 2, a diffusion model capable of creating 3D worlds and sustaining them for longer. The model generates images as the player moves through the simulating world, inferring ideas about the environment and handling first-person and isometric viewpoints.

Genie 2 can remember parts of a simulated scene even after they leave the player's field of view. However, it can generate consistent worlds for up to 60 seconds.

Read the full story here

2. Luma Photon and Photon Flash - the most creative, intelligent and personalizable image generation models

Luma Photon and Photon Flash are new image generation models built on a groundbreaking architecture that delivers ultra high quality and 10x higher cost efficiency. These models are designed to be creative, intelligent, and personalizable, allowing designers, movie makers, architects, and visual thinkers to explore vast idea spaces and achieve extraordinary things.

In large-scale double-blind evals, Luma Photon outperforms every model on the market in quality, creativity, and understanding while being radically more efficient. The Luma Dream Machine service is built on this principle of visual abundance.

Read the full story here

3. Safeguarding AI against ‘jailbreaks’ and other prompt attacks

Microsoft is addressing prompt attacks on AI tools and applications by implementing a comprehensive approach to detect, measure, and manage risk. The company has developed Prompt Shields, a model for real-time blocking of malicious prompts, and safety evaluations for simulating adversarial prompts.

Microsoft Defender for Cloud helps prevent future attacks, while Microsoft Purview manages sensitive data used in AI applications. The company also publishes best practices for developing a multi-layered defense that includes robust system messages and rules guiding AI model safety and performance.

Read the full story here

4. Here's What OpenAI's $200 Monthly ChatGPT Pro Subscription Includes

OpenAI has launched ChatGPT Pro, a $200-per-month subscription tier that offers unlimited access to all of its models, including the full version of its o1 "reasoning" model. The subscription aims to cater to power users of ChatGPT who are already pushing the models to the limits of their capabilities on tasks like math, programming, and writing.

The new version of o1 is generally more performant, faster, more powerful, and accurate, and can reason about image uploads. However, the full o1 performs worse on common benchmarks, such as MLE-Bench.

Read the full story here

5. Microsoft’s Copilot can now browse the web with you using AI ‘Vision’

Microsoft is testing its new Copilot Vision feature, which allows its AI companion to read webpages in Microsoft's Edge browser. Users can ask questions about text, images, and content, or use it to assist them. Copilot Vision is an optional experience, and users must grant permission for it to read webpages.

The feature is currently limited to Copilot Pro subscribers through Microsoft's Copilot Labs program. Privacy concerns are expected as AI models start reading web content, and Microsoft is taking time to roll out the feature. Copilot Vision will only interact with a select set of websites initially, with plans to expand access to more Pro subscribers and websites over time.

Read the full story here