📢 Pyramid Flow, New Computationally Efficient Open Source Video Generator

A new open-source video generation tool joins the fray, claiming efficiency gains.

Author

Jack Lajoie
October 10, 2024 • read time ~ 7 minutes

The Daily Current ⚡️

We have a new open-source video model to play with! It’s October 10th, and AI news continues to hum right along.

The Authors Guild is taking a stand against AI’s effect on their industry, Wall Street declares the AI experimentation honeymoon to be over, and Microsoft's healthcare AI tools promise to revolutionize patient care. Finally, Meta’s new Advantage ad suite promises to automate everything from creative to audience targeting.

 đź”Ś Plug Into These Headlines:

  1. New Open Source Video Generator: Pyramid Flow

  2. Authors Guild fights back against AI with “Human Authored” label

  3. The honeymoon is over for AI experimentation

  4. Microsoft's healthcare AI push deepens

  5. Meta publishes new overview of its AI-based “Advantage” ad suite

Pyramid Flow joins the growing list of high quality AI video generation tools, and it’s open-source. It’s able generate high-quality 10-second videos at 768p resolution with 24 FPS, and only trained on open-source datasets.

  • “training-efficient” autoregressive video generation method based on Flow Matching

  • Built on the SD3 architecture with 2 billion parameters

  • Utilizes flow matching techniques for efficient video generation

  • Trained on open-source datasets within 20.7k A100 GPU hours

You can check out their project page and view samples here.

In an effort to protect human creativity, the Authors Guild has introduced the "Human Authored" label, a certification for books written by actual humans— not AI. This move is a response to the growing concern over AI-generated content flooding the literary world.

  • The Human Authored label aims to assure readers of authentic, human-crafted stories.

  • The initiative follows a broader effort to counteract AI’s rise in creative spaces, including copyright lawsuits against AI companies.

  • No initial use of AI detection software due to current limitations in accuracy

  • System relies on:

    • Self-reporting by authors

    • Community vigilance to identify potential fraud

  • Authors must certify their work as human-written at risk of legal action for false claims.

At the WSJ’s CIO Network Summit, business leaders are reportedly looking for “better uses of AI than just chatbots”. Many companies that enthusiastically invested in AI projects are now being pressured to prove those investments are paying off. What started as a playground of innovation is now a race to deliver measurable returns.

  • As budgets tighten, profitability and productivity are key metrics demanded by investors.

  • Initial AI projects, like automated customer service and predictive analytics, were widely adopted, but scalability issues persist.

  • Executives now face the challenge of demonstrating that AI isn’t just hype but a reliable driver of business value.

The pressure is mounting as stakeholders want to see tangible returns on these AI initiatives.

Microsoft just announced a major expansion in its AI-driven healthcare tools, further embedding its Azure platform into the sector. This includes models designed for medical imaging and data interoperability, among others.

  • The highlight is CXRReportGen, which can generate chest X-ray reports for physicians to review and augment.

  • Microsoft’s Fabric platform now offers healthcare-specific data solutions, aiming to overcome interoperability challenges across disparate systems.

  • A new customizable healthcare service agent integrates with Microsoft Copilot, offering personalized automation for patient care.

With competition from Google’s Med-PaLM and Amazon’s AWS, Microsoft is likely seeking to solidfy its place in the healthcare AI race

Meta has been pushing its AI-based Advantage ad suite, after showcasing its latest updates at the Agency Summit. These tools offer advertisers a new level of control—or lack thereof—as many are finding out that Meta’s AI does a lot of the heavy lifting for them, from creative generation to targeting and bidding.

  • Audience expansion capabilities help uncover potential customers that traditional targeting may miss.

  • Creative and messaging are now customized automatically to fit different user groups.

  • Some advertisers may feel nervous about giving up too much control, but results are reportedly strong across the board.

Meta looks to have quietly built the automated AI marketing agent many dreamed of this year. However it’ll of course only work within Meta’s platforms.