📰 News 📰 

EMO: Generating Expressive Portrait Videos

The EMO project, developed by the Institute for Intelligent Computing at Alibaba Group, introduces an innovative audio-driven portrait-video generation framework. This technology utilizes a single reference image and vocal audio to create vocal avatar videos with expressive facial expressions and various head poses. The framework supports multiple languages and maintains character identity over any video duration. The demo’s on their site are quite impressive and puts EMO into the video generation race.

I highly recommend checking out the examples on their Github

Pika Labs Innovates with Lip Sync for AI-Generated Characters

Continuing on the video generation news. Pika Labs has introduced a groundbreaking Lip Sync feature to its AI video generation platform, allowing AI-generated characters to speak with synchronized lip movements. Developed alongside AI audio platform ElevenLabs, this feature marks significant progress in AI video technology, making it possible for characters to have realistic conversations. This advancement is particularly beneficial for filmmakers who previously relied on dubbing for character dialogue. The Lip Sync feature is exclusive to Pika Labs' Pro plan subscribers, priced at $58 per month. Despite not being perfect, it represents a significant step forward in AI-driven content creation. The AI video generation industry is poised for significant growth in 2024, with companies like Runway ML, Stable Diffusion, and OpenAI's Sora expanding their offerings.

Read More At: PetaPixel

OpenAI's Legal Struggles: Further Allegations Unfold

OpenAI accuses The New York Times of paying someone to 'hack' ChatGPT to generate verbatim paragraphs from its articles. The NYT sued OpenAI and Microsoft for scraping its website without permission, claiming evidence of ChatGPT reproducing whole passages.

And on a separate front, three digital publishers, including The Intercept, Raw Story, and AlterNet, have filed lawsuits against OpenAI for using their copyrighted articles to train ChatGPT without permission. The media organizations accuse OpenAI of copyright infringement and violating the Digital Millennium Copyright Act.

💬 Large Language Models 🗨️ 

Nvidia, Hugging Face, and ServiceNow Unveil StarCoder2

Nvidia, Hugging Face, and ServiceNow have launched StarCoder2, an advanced open-access code generation LLM available in three sizes. Trained on over 600 programming languages, StarCoder2 aims to enhance developer productivity and is part of the open BigCode Project. The models offer scalable solutions for enterprises, promoting responsible AI development and usage under OpenRAIL licenses. This collaboration signifies a significant step towards efficient, responsible AI-driven code generation.

Read More At: VentureBeat

Efficient Adversarial Prompt Generation for Language Models

Computer scientists have developed BEAST, a fast method to create harmful prompts for language models using an Nvidia GPU. This technique outperforms gradient-based attacks, achieving an 89% success rate in just one minute. BEAST can be used to elicit inaccurate responses, conduct privacy attacks, and improve existing toolkits. Despite its effectiveness, thorough safety training can mitigate its impact, emphasizing the importance of ensuring AI model safety for future deployments.

Read More At: The Register

Challenging AI Norms: Introducing Antagonistic AI

Researchers from MIT and the University of Montreal propose Antagonistic AI, challenging the overly-sanitized behavior of current large language models. They argue that AI systems should be combative, critical, and even rude to promote resilience, personal growth, and diversification of ideas. By implementing techniques like opposition, personal critique, and violating interaction expectations, antagonistic AI can provide entertainment while fostering self-reflection and enlightenment. The researchers emphasize the importance of building responsible antagonistic AI that respects consent, context, and framing.

Read More At: VentureBeat

🔨 AI Powered Tools 🔨 

A run down of the latest SaaS products and services which leverage AI to help you take back some time.

• Descript - Create high-quality audio just by typing. Generate your own voice clone or choose from their stock AI voices

• Runway - Media creation tools. Everything you need to make anything you want from image generation to video editing.

• Personal AI - Private models trained on your data to create an individual experience for each user

• Copilot for Finance - Microsoft introduces Copilot for Finance, an AI assistant for finance professionals, leveraging Excel and ERP data to automate tasks and streamline financial operations.

🗞️ Other News Rollup 🗞️ 

AI-Powered AR Tool - Shader, led by ex-Snap designer, offers easy AR creation with AI, plans premium features and social sharing.

Music Generation AI - Adobe's Project Music GenAI Control enables customizable music creation from text or melodies, raising ethical and legal concerns.

AI Event Disaster - Glasgow's Willy Wonka AI event disappoints attendees with sad props and false promises. Refunds offered.

EU Edge AI - European consortium PREVAIL prototypes edge AI tech to boost Europe's position in the global market.

AI Against Scam Calls - Microsoft's AI tool detects and alerts on suspicious calls to combat spam and scams.

Humanoid Robotics Challenges - Figure AI, backed by Bezos, OpenAI, and Nvidia, faces high risk in developing humanoid robots for various tasks.

AI Database Evolution - Couchbase adds vector search for AI, expanding capabilities for adaptive applications in cloud, mobile, and edge computing.

AI Governance Solution - Collibra's AI Governance ensures safe, efficient AI deployments with risk mitigation strategies, enhancing transparency and model performance.

🎶 Prompts 🎶 

Prompts used to generate some of this issues images. Unless otherwise stated all images are using Dall-E and ChatGPT Plus.

That’s all for today.

So what did you think?