Google’s Quantum Leap 🌌 Sora Finally Lands 🎥 Meta Goes Nuclear 🚀
PLUS: 12 days of OpenAI, GenCast predicts the impossible, and Sora makes waves in video generation.
👋 Welcome to this week in AI.
🎵 Don’t feel like reading? Listen to two synthetic podcast hosts talk about it instead.
📰 Latest news
Willow: Google’s Quantum Breakthrough That Could Redefine the Future
Google has introduced Willow, its latest quantum processor, showcasing major advancements in quantum computing.
Willow completed Random Circuit Sampling (RCS), a task that would take classical supercomputers over 10 septillion years, in just five minutes.
10 septillion year = 100,000,000,000,000,000,000,000,000 years
This achievement demonstrates the processor's ability to tackle problems previously beyond the reach of traditional computing.
One of Willow’s key innovations lies in error correction. Unlike earlier processors, Willow improves its performance as the number of qubits increases, addressing a fundamental challenge in quantum scalability.
With 105 qubits and coherence times up to 100 microseconds—five times longer than its predecessor, Sycamore—Willow shows promise in addressing real-world applications like drug discovery, battery design, nuclear fusion, and logistical optimisation.
Why It Matters
Willow’s ability to solve problems at speeds unattainable by classical computers illustrates quantum computing’s potential to reshape industries.
Its advancements in error correction bring quantum systems closer to practical use by addressing scalability, a longstanding barrier in the field.
Quantum computing could render current cryptographic systems obsolete, forcing the development of quantum-resistant encryption to safeguard sensitive data.
In healthcare and material science, its potential to transform drug discovery and battery innovation could lead to breakthroughs in treatments and energy storage.
Moreover, applications in nuclear fusion and logistics optimisation promise to enhance industries ranging from energy production to global supply chains.
Willow’s progress underscores the growing possibilities of quantum computing to address complex, real-world problems, paving the way for advancements that could redefine technology and society.
12 Days of OpenAI: Sora’s Long-Awaited Launch and Pro Tools for Power Users
OpenAI’s “12 Days of OpenAI” event highlights a series of daily product updates and demonstrations, showcasing the company’s latest advancements in AI.
The event kicked off with significant releases and has maintained momentum with notable reveals each day.
Key releases so far include:
Day 1 - o1 Reasoning Model:
A new reasoning model offering 34% fewer errors compared to the preview version.
Enhanced capabilities for coding, maths, and writing.
Available to ChatGPT Plus and Pro users, with Enterprise and Education support coming soon.
Achieves 80% reliability in technical benchmarks, significantly improving problem-solving.
Day 2 - ChatGPT Pro Subscription:
A premium $200/month tier aimed at power users with high computational demands.
Features include a 128k context window and access to enhanced versions of o1 for complex tasks.
Focused on reliability for technical and professional applications.
Day 3 - Sora Video Generator:
A text-to-video model enabling 20-second video creation at resolutions up to 1080p.
Includes remixing tools and a storyboard editor for precise control.
Available to ChatGPT Plus and Pro users, with Pro offering expanded credits and higher output quality.
Early features like metadata tagging and watermarks enhance transparency and safety.
Day 4 - ChatGPT Canvas:
A side-by-side document editor integrated with ChatGPT for real-time collaboration and Python execution.
Allows users to edit, run code, and troubleshoot directly within the interface.
Why it Matters
Sora’s release marks a pivotal moment in video generation. After a over year of anticipation, finally OpenAI has released it. Sam Altman compared it to the release of GPT-3, indicating that the model will get remarkably better over time.
ChatGPT Pro’s $200/month subscription represents a significant shift in the market for large language models.
While expensive compared to other subscriptions, it delivers unique advantages for academia, research, and technical professionals. Features like a 128k context window and enhanced reliability make it invaluable for tackling complex challenges, from scientific modelling to advanced data analysis.
OpenAI continues to innovate and push the boundaries of what’s possible.
📝 The 12 days of OpenAI release page
📝 Try SORA (if you have a ChatGPT Plus subscription, you can use it now)
Copilot Vision: New Digital Species
Microsoft’s Copilot Vision introduces real-time screen understanding and conversational AI, transforming how users interact with technology.
Integrated into the Edge browser, it dynamically adapts to user actions and emotions, offering personalised and emotionally intelligent responses.
Privacy-focused features include session-based data deletion, while future updates promise persistent memory, task automation, and gaming integration.
Microsoft envisions Copilot Vision as a digital companion that seamlessly integrates into daily life, guiding tasks and understanding user preferences.
Why It Matters
Copilot Vision redefines user interaction by replacing traditional interfaces with conversational, emotionally aware engagement. Its privacy-centric design, including session-based data deletion, builds trust while laying the groundwork for secure memory storage in the future.
By integrating deeply into tools like browsers and gaming platforms, Microsoft positions Copilot Vision as an indispensable companion for productivity, learning, and entertainment.
This marks a critical step toward creating AI that feels personal and intuitive, balancing functionality with user-centric design.
📝 Read more on Microsoft's blog
Genie 2: Redefining Gaming and AI with Interactive 3D Worlds
Check out the gameplay ^^^
DeepMind’s Genie 2 brings transformative potential to gaming and AI research, enabling the generation of interactive 3D environments from simple text or images.
With advanced features like realistic physics, long-horizon memory, and dynamic animations, Genie 2 supports diverse applications, from creating immersive game worlds to training AI agents in complex scenarios.
Its ability to render consistent, playable environments for up to a minute showcases its potential as a tool for prototyping and creativity.
Why It Matters
Genie 2 opens up new possibilities for gaming by allowing developers to rapidly create interactive worlds with minimal input.
This tool could revolutionise game design by reducing the time and resources needed to develop complex environments, while also enabling endless customisation and experimentation.
For AI development, Genie 2 provides diverse, dynamic settings for training agents, fostering adaptability and improving task performance. Its ability to simulate realistic physics and interactions positions it as a valuable resource for advancing embodied AI research.
By blending creativity and technical precision, Genie 2 has the potential to reshape both gaming and AI research, paving the way for innovative experiences and more generalised AI systems.
📝 Blog by Google (lots of examples)
Meta Goes Nuclear: A Sustainable Future for AI
Meta has announced an initiative to support nuclear energy development, targeting 1-4 gigawatts (GW) of new nuclear generation capacity in the U.S. by the early 2030s.
To put this into perspective, 4 GW would equate to approximately 13.2% to Australia’s average annual electricity generation.
This ambitious project is part of Meta’s strategy to meet the escalating energy demands of AI innovation while ensuring the sustainability of its data centres.
By issuing a request for proposals (RFP), Meta plans to collaborate with nuclear developers to scale the technology, reduce costs, and navigate regulatory and operational challenges.
Building on a decade of renewable energy investments, including 12,000 MW in solar, wind, and geothermal projects, Meta is turning to nuclear energy for its reliability and longevity.
This move aligns with similar efforts by other major tech companies, such as Microsoft and Google, which are also exploring nuclear power as a solution to the energy-intensive requirements of AI advancements.
Why It Matters
Meta’s focus on nuclear energy is critical for addressing the surging power demands driven by AI innovation and data centre expansion.
With nuclear providing a stable and scalable energy source, Meta positions itself to support its growing operational needs while maintaining reliability.
By collaborating with developers, Meta aims to accelerate nuclear technology, lower costs through scaling, and address the regulatory complexities inherent in such projects.
These efforts mirror a broader trend in the tech industry, with companies like Microsoft and Google similarly turning to nuclear energy to meet their energy-intensive AI ambitions.
GenCast: A New Era of Weather Forecasting
DeepMind has unveiled GenCast, an AI-powered weather forecasting system that sets a new benchmark for precision, efficiency, and accessibility.
By outperforming the European Centre for Medium-Range Weather Forecasts model (ENS) on 97% of evaluation metrics, GenCast delivers unmatched accuracy, providing reliable predictions essential for sectors like agriculture, transportation, and disaster response.
Its ability to process 15-day forecasts in just 8 minutes on a single AI chip eliminates the need for costly supercomputers, making advanced forecasting more accessible.
The system excels at predicting extreme weather events, including tropical cyclones and heat waves, which can save lives and mitigate economic losses by enabling faster responses.
Trained on 40 years of historical weather data, GenCast also supports open collaboration by making its code available for non-commercial research, fostering global innovation in meteorology and climate resilience.
Why It Matters
GenCast’s exceptional accuracy provides reliable predictions, offering critical insights that enhance decision-making across industries such as agriculture, transportation, and disaster response.
By processing 15-day forecasts in just 8 minutes, the system reduces reliance on traditional supercomputers, cutting costs and making advanced forecasting tools more accessible.
Its ability to predict extreme weather events like cyclones and heat waves strengthens disaster management efforts, protecting lives and reducing economic losses.
Additionally, the open-sourcing of GenCast fosters collaboration, encouraging researchers worldwide to advance weather forecasting technology and unlock new applications across diverse fields.
