Good morning. It’s Wednesday, October 16th.

Did you know: Concepts of the self-operating machine date back to the 16th century?

In today’s email:

• Open-Source AI Video Generation

• Adobe Firefly Video

• Full-body Motion Capture App

• OpenAI Strengthens Its Ranks

• Google Goes Nuclear

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Daily News for Curious Minds

“I stopped watching the news, so sick of the bias. Was searching for an alternative that would just tell me WHAT happened, with NO editorializing. I found it. It’s called 1440. It assumes you are smart enough to form your own opinions.”

Sign up now!

New AI model for hi-res video generation, Pyramid Flow, is available as open-source software

A team of researchers has introduced Pyramid Flow, an open-source AI model for generating high-resolution (768p) video imagery. This model provides a cost-effective alternative for creating virtual video content, bypassing the need for traditional filming. Pyramid Flow utilizes a multi-stage generation process, producing videos in lower resolutions before reaching the final output.

The inference shell can create a five-second video in 56 seconds at 384p resolution while using significantly less computing power and minimizing token requirements for generation. The developers have released the model's code on GitHub under an MIT License, along with sample videos demonstrating its performance and the open-source datasets utilized for training, totaling 10 million short videos. Read more.

Adobe Firefly Video: Generative AI Video Model 

Adobe has begun rolling out its Firefly Video Model, an AI tool that generates video from text prompts, entering a competitive space alongside OpenAI’s Sora and Meta’s recent video AI efforts. While major players like ByteDance and Meta have released similar tools, Adobe is distinguishing itself by training models on legally cleared data, ensuring that commercial use is hassle-free.

Although a general release date for the tool is still pending, Adobe has opened access to users on its waitlist. PepsiCo's Gatorade and Mattel’s Barbie packaging designs are already utilizing Adobe's image AI, signaling practical applications for its technology.

Adobe's focus with Firefly is on fine-grain control, offering video creators tools that integrate seamlessly with traditional footage. This positions Adobe to cater to everyday users, especially video editors, by enabling precise control over elements like camera motion and angles.

Alongside the Firefly Video Model, Adobe has also enhanced its Photoshop tools and introduced new features in Illustrator and InDesign, further solidifying its commitment to advancing AI capabilities within its Creative Cloud suite. Read more.

New app performs real-time, full-body motion capture with a smartphone

Northwestern University engineers have introduced MobilePoser, a new app that performs real-time, full-body motion capture using sensors already embedded in consumer devices like smartphones, smartwatches, and wireless earbuds.

By utilizing inertial measurement units (IMUs) and advanced AI algorithms, MobilePoser accurately estimates joint positions, rotations, and walking speed without requiring specialized equipment. A physics-based optimizer further enhances motion prediction accuracy.

With a tracking error of 8 to 10 centimeters, the system adapts to various device setups and has potential applications in gaming, fitness, and healthcare, allowing for improved tracking of mobility and posture analysis. Read more.

OpenAI Strengthens Its Ranks

OpenAI has landed two strategic hires after a mass exit earlier this month. First up is Sebastian Bubeck, previously Microsoft’s VP of generative AI research, credited with shaping the Phi models that power efficient on-device AI through smaller language and vision frameworks. As the appetite for nimble, privacy-first models surges, Bubeck is set to tackle efficiency and model development—areas where OpenAI has lagged behind.

Joining him is Dane Stuckey, ex-CISO of Palantir, who brings a wealth of experience in digital forensics and incident response. Stuckey’s new role coincides with OpenAI’s push to strengthen its security infrastructure, particularly in its partnerships with the U.S. Department of Defense. OpenAI recently lifted its ban on selling AI technology to the military and has been working on various projects with the Pentagon, including cybersecurity initiatives. Read more.

Google signs deal with nuclear company as data center power demand surges

Google has partnered with Kairos Power to address the rising power demands of its data centers, which are increasingly reliant on clean, reliable energy. Under the agreement, Google will source power from Kairos' small modular reactors (SMRs), the first of which is projected to be online by 2030, with 500 megawatts added to the grid by 2035.

Kairos' reactors, using molten fluoride salt instead of traditional water coolant, offer a more scalable and potentially cost-effective nuclear solution. This move aligns with a broader trend of tech companies, such as Microsoft and Amazon.

Nuclear energy, offering round-the-clock, emissions-free power, is seen as vital for meeting the industry's growing needs without compromising on sustainability targets. Read more.

• ComfyGen AI automates multi-stage text-to-image workflows from simple prompts

• Physicists uncover behavior in quantum superconductors that provides a new level of control

• AI chatbots need some healthy disagreement to work best together, tech exec says

• AI model simulates Counter-Strike with 10 FPS on a single RTX 3090

• OpenR: An Open-Source AI Framework Enhancing Reasoning in Large Language Models

• US mulls capping Nvidia AI chips exports to some countries, Bloomberg News reports

• Nvidia just hit another record high

• Domino Data Lab seeks to embed governance in AI development

• Anthropic just made it harder for AI to go rogue with its updated safety policy 

• Alibaba's international arm says its new AI translation tool beats Google and ChatGPT

• Google supercharges Shopping tab with AI and personalized recommendation feed

• Cognizant adds multi-agent functionality to AI application platform 

• The New iPad Mini Gets an Apple Intelligence Boost

• OpenAI says ChatGPT has much less gender bias than all of us

• Meta researchers develop method to make AI models "think" before answering  

• How Tesla's plans for 'unsupervised FSD' and robotaxis could run into red tape 

• YouTube rolling out new miniplayer, fine-tunable playback speed, sleep timer, and more 

• EU AI Act checker reveals Big Tech's compliance pitfalls

• NYT sends AI startup Perplexity 'cease and desist' notice over content use

📄 From Generalist to Specialist: Adapting Vision Language Models via Task-Specific Visual Instruction Tuning

📄 LOKI: A Comprehensive Synthetic Data Detection Benchmark using Large Multimodal Models  ​​

📄 Animate-X: Universal Character Image Animation with Enhanced Motion Representation  

📄 MTU-Bench: A Multi-granularity Tool-Use Benchmark for Large Language Models

📄 MMIE: Massive Multimodal Interleaved Comprehension Benchmark for Large Vision-Language Models

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!