Adobe Launches Firefly Video AI

Sign Up | Advertise

The AI World Today

• Adobe Launches Firefly Video AI

• Google Signs Nuclear Power Deal

• UK Secures Major AI Investments

  +

• Heads Up

• AI Solution

Adobe Unveils Firefly Video Model and Tools

Adobe has launched a public test of its Firefly Video Model, allowing users to create videos from text prompts and integrate features like camera angles, motion, and style. Available on the Firefly website, the tool is part of Adobe’s broader effort to provide AI-powered creative tools, with content generated using licensed materials for commercial use. Additionally, Adobe has introduced a generative extend feature in Premiere Pro, enabling creators to lengthen existing video clips. The video model is accessible only to paid Creative Cloud customers and includes safety measures, such as prohibiting videos with children or public figures. Videos generated with Firefly will have embedded content credentials indicating AI involvement. These updates underscore Adobe’s commitment to enhancing AI-driven video creation.

Google Partners with Kairos for Nuclear Power

Alphabet's Google has signed the world’s first corporate agreement to purchase power from small modular reactors (SMRs) to meet growing electricity demand for artificial intelligence. The deal with Kairos Power aims to bring its first reactor online by 2030, with plans to deploy six to seven reactors by 2035, generating 500 megawatts of power. Unlike traditional reactors, SMRs are smaller and built with factory-made components to reduce construction costs. This follows recent tech industry moves towards nuclear energy, including agreements by Amazon and Microsoft. The power from these reactors will help meet the rising energy needs of AI and data centers, which are projected to triple by 2030.

American Tech Firms Invest Billions in UK Data Centers

The U.K. government announced securing £6.3 billion ($8.23 billion) in data center investments from four U.S. tech companies to support AI-driven demand. Investment pledges came from CyrusOne, ServiceNow, CloudHQ, and CoreWeave. CyrusOne will invest £2.5 billion in U.K. projects by 2028, ServiceNow plans to expand its London and Newport facilities with a £1.15 billion investment over five years, and CloudHQ will develop a £1.9 billion data center campus in Oxfordshire. CoreWeave will contribute £750 million to AI cloud infrastructure. This follows a broader trend of U.S. tech companies expanding globally, with firms like Amazon and Microsoft committing billions to cloud and AI infrastructure across Europe to meet growing technological demands.

Heads Up

Zyphra has launched Zamba2-7B, a cutting-edge 7B-scale language model outperforming Mistral, Google’s Gemma, and Meta’s Llama3 in quality and performance, optimized for on-device and enterprise use.

OpenAI is in a trademark dispute with Guy Ravine, who owns the ‘Open AI’ trademark and claims to have pitched the idea before the company’s founding. 

Microsoft AI VP Sebastien Bubeck is leaving after 10 years to join OpenAI, where he will focus on advancing efforts towards artificial general intelligence (AGI). 

Deloitte reports that the global semiconductor industry still needs to add one million skilled workers by 2030, as it races towards $1 trillion in revenue.

China's registered users of large generative AI models have surpassed 600 million, according to the Ministry of Industry and Information Technology.

AI Solution

DIAMOND: Diffusion-Based Breakthrough in Reinforcement Learning

Researchers have introduced DIAMOND (DIffusion As a Model Of eNvironment Dreams), a cutting-edge reinforcement learning agent trained using a diffusion world model. Unlike traditional models that rely on discrete latent variables, DIAMOND leverages diffusion to capture visual details crucial for environment dynamics, improving agent performance. This shift in approach follows the growing dominance of diffusion models in image generation. DIAMOND achieves a new best mean human normalized score of 1.46 on the competitive Atari 100k benchmark, setting a record for agents trained entirely within a world model. The researchers have released the code, agents, and playable world models, fostering further exploration of diffusion techniques in world modeling. The project is available on GitHub at https://github.com/eloialonso/diamond.