Superpower Daily
Posts
Meta AI strikes again with highly realistic head avatars

Meta AI strikes again with highly realistic head avatars

Google’s best Gemini demo was faked

Saeed Ezzati
December 08, 2023

In today’s email:

🤯 Screenshot to Code with v0 is crazy
🙏🏼 Anthropic’s latest tactic to stop racist AI: Asking it ‘really really really really’ nicely or someone will sue us
👾 A universal backdoor for poisoning large image classification models
🧰 10 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

Screenshot to Code with v0 is crazy

Here's just a simple screenshot of Twitter being recreated in code with AI in a matter of seconds.

Soon anyone will be able to create software.

v0 is a generative user interface system developed by Vercel, designed to create React code using AI models. It allows users to generate user interfaces from text prompts, offering three choices of AI-generated UIs that can be further refined and integrated into projects. This system uses a combination of custom code, open-source, and synthetic datasets for training, but explicitly excludes Vercel customer data.

v0 currently supports React DOM elements and shadcn/ui components, with plans to expand to other UI libraries and design systems. It's in Private Beta, accessible via a waitlist, focusing on frontend code generation with potential future capabilities for data fetching code.

News. Without Motives. That’s 1440. Join Today!

Frustrated by biased news? Tired of trying to sift through 100s of news sources to find the facts? Leave behind the opinions and receive a daily digest of pure facts. Say hello to 1440! Delve into politics, business, science, sports, culture, and more. Join the other 3 million Americans who read 1440 each morning.

Meta AI strikes again, with Relightable Gaussian Codec Avatars

In their latest work, META AI researchers have developed "Relightable Gaussian Codec Avatars," a groundbreaking method for creating highly realistic, relightable head avatars. These avatars are capable of displaying new expressions with remarkable fidelity. The core of this technology lies in a geometry model based on 3D Gaussians, which accurately captures intricate details such as individual hair strands and skin pores in dynamic facial sequences.

To accommodate the varying materials found in human heads, like eyes, skin, and hair, the team has introduced a novel relightable appearance model. This model leverages learnable radiance transfer techniques and incorporates global illumination-aware spherical harmonics for managing diffuse components. The result is an avatar that can be realistically relit in real time under various lighting conditions, including both point light sources and continuous illumination, with comprehensive all-frequency reflections.

A notable advancement in this technology is the enhanced fidelity of eye reflections and the ability to control gaze direction through relightable explicit eye models. These features add to the avatar's lifelike appearance and interaction capabilities.

Building with OpenAI – ChatGPT, Agents, and the Path Towards AGI

with OpenAI’s Romain Huet, Head of Developer Experience

Just one year ago, the world underwent a seismic shift as ChatGPT, the pioneering generative AI chatbot, made its debut. In an instant, the prospect of a science-fiction future where machines possess autonomous thought felt more real than ever. One year later, OpenAI brings us one step closer to where we are headed with the debut of GPTs and their Assistants API, which makes it easy for developers to create assistive AI apps that have goals and can call models and tools.

OpenAI’s Romain Huet, Head of Developer Experience, will talk about this first step towards helping developers build agent-like experiences within their own applications, his insights and advice on the opportunities of building on the Assistants API, and what’s possible. Guiding this conversation is Vince Hankes, a Partner at Thrive Capital.