🤯Google Goes Nuclear for AI

Google goes nuclear to meet AI’s growing power needs

Key Points 

• Google has signed a landmark deal to source energy from small nuclear reactors to support the increasing demands of AI.

• The agreement includes six or seven reactors from Kairos Power, with the first expected to be operational by 2030.

• This deal highlights trust in small modular reactor (SMR) technology, designed to reduce costs and construction delays typical of larger plants.

☕News - Google has signed a landmark agreement to source energy from a fleet of small nuclear reactors to meet the growing energy needs of AI — a move they believe offers a clean and reliable solution to their electricity needs around the clock.

The company has ordered six or seven reactors from Kairos Power in California, with the first one expected to be operational by 2030 and the rest by 2035. This deal is part of Google's plan to power its data centers with low-carbon energy. 

📊A new trend? The rapid rise of generative AI and cloud storage has driven up the energy demands of tech companies. In response, Microsoft recently signed a deal to source power from Pennsylvania's Three Mile Island, marking the nuclear plant’s first activation in five years since its 1979 reactor meltdown. Similarly, in March, Amazon purchased a nuclear-powered data center in Pennsylvania from Talen Energy.

🥸Insider's opinion - Michael Terrell, Google's senior director for energy and climate, underscored the need for new power sources to support AI technologies that are advancing science, improving business services, and driving economic growth. He noted that the agreement is a step toward providing clean and reliable energy, helping unlock AI’s full potential.

The deal, which still requires regulatory approval, shows confidence in small modular reactor (SMR) technology. These smaller, factory-built reactors are designed to avoid the cost overruns and delays of larger plants. However, critics argue SMRs might still be costly since they may not achieve the same economies of scale as bigger reactors.

Adobe kicks off rollout of Firefly video tools, taking on OpenAI and Meta

Key Points 

• Adobe’s Generative Extend tool allows users to extend video clips by up to two seconds and correct minor editing issues.

• The new Text-to-Video and Image-to-Video tools enable users to create videos directly from text descriptions or reference images.

• Adobe's tools are considered commercially safe, trained on permitted content, distinguishing them from competitors like Runway and Meta.

🎥News - Adobe is entering the generative AI video space with the launch of its Firefly Video Model, which has been in development since earlier this year. This new feature is now integrated into various tools, including Premiere Pro, enabling users to extend video footage and create videos from still images and text prompts.

The first tool, Generative Extend, is in beta for Premiere Pro and allows users to extend the beginning or end of clips that may be too short, as well as make adjustments during a shot to correct minor issues like eye-line shifts or unexpected movements. It can extend clips by a maximum of two seconds, making it ideal for small adjustments. 

Extended clips can be produced in either 720p or 1080p at 24 FPS. The tool also offers some audio editing capabilities, allowing users to extend sound effects and ambient “room tone” by up to ten seconds, though it does not extend spoken dialogue or music.

✨What's more? Adobe is also launching two new video generation tools as a limited public beta in the Firefly web app: Text-to-Video and Image-to-Video. These tools, first announced in September, allow users to create videos directly from text or images.

Text-to-Video allows users to input a text description to generate videos, similar to other platforms like Runway and OpenAI's Sora. It can produce clips in various styles, including traditional film, 3D animation, and stop motion. Users can refine their creations using camera controls that simulate angles, motion, and shooting distance.

Image-to-Video enhances this process by letting users upload a reference image along with a text prompt, offering more control over the output. This feature is useful for creating b-roll or visualizing reshoots with stills from existing footage. 

Currently, both tools can generate clips up to five seconds long, with a maximum quality of 720p at 24 frames per second. Each clip takes about 90 seconds to create, but Adobe is developing a “turbo mode” to speed up the process.

🔥The Adobe advantage - Adobe's tools, despite certain restrictions, are “commercially safe” since they’re trained on content the company has permission to use. This sets them apart from other providers like Runway, which are under scrutiny for potentially training their models on thousands of scraped YouTube videos. So if commercial viability is your concern, Adobe's got that covered.

Another advantage of Adobe’s Firefly video model is that videos created or edited with it can include Content Credentials, which disclose AI usage and clarify ownership rights when published online. While it’s unclear when these tools will move out of beta testing, at least they are currently publicly accessible—unlike OpenAI’s Sora, Meta’s Movie Gen, and Google’s Veo generators.