## The Dawn of Interactive Video
In a world where technology continually reshapes our experiences, London-based AI lab Odyssey has launched a groundbreaking research preview that promises to redefine entertainment through AI-generated \1. Initially aimed at enhancing film and game production, the Odyssey team has unearthed a potential new medium that invites users to step into a realm where video transforms into an interactive playground.
The Odyssey model creates an experience akin to what Star Trek fans might call the “Holodeck.” This immersive \1 allows users to interact with video content through various devices, including keyboards, game controllers, and voice commands, making the experience highly responsive. Each interaction is met with the generation of realistic-looking video frames every 40 milliseconds, enabling users to feel as though they are actively influencing the digital environment. As Odyssey describes it, “the experience today feels like exploring a glitchy dream—raw, unstable, but undeniably new.” This description aptly captures the nascent stage of this technology, which, while not yet delivering the polished visuals of AAA games, teases the incredible potential that lies ahead.
## Understanding the Technology Behind the Magic
To comprehend what makes Odyssey’s \1 so distinct, we must delve into the concept of “world models.” Unlike traditional video formats that produce complete clips, world models operate on a frame-by-frame basis, predicting subsequent frames based on the current state and user inputs. This approach mirrors the mechanics of large language models that predict the next word in a sentence, yet it operates on a far more intricate level, generating high-resolution video instead of mere text.
\1 characterizes a world model as “an action-conditioned dynamics model.” Each user interaction feeds into the model, which utilizes the current state, the action taken, and historical context to produce the next frame. This framework encourages a more organic and unpredictable experience, breaking away from the rigid programming found in conventional games, where specific actions lead to predetermined outcomes. Instead, the AI must rely on its extensive video training to make educated guesses about what should occur next.
## Overcoming Historical Challenges in AI-Generated Video
Creating something as ambitious as \1 is fraught with challenges. A significant obstacle is maintaining stability over time; generating each frame based on previous ones can lead to compounding errors—an issue commonly referred to as “drift” among AI researchers. In response, Odyssey has developed what they term a “narrow distribution model,” pre-training their AI on a broad spectrum of video footage before fine-tuning it within a limited set of environments. While this approach sacrifices some variety, it enhances stability, preventing the experience from devolving into chaos.
Currently, \1 reports they are making rapid strides towards a next-generation model that promises to deliver a richer array of pixels, dynamics, and actions. However, the technological prowess required to run such advanced AI in real-time comes at a cost. Presently, the infrastructure supporting this innovative experience incurs expenses between £0.80-£1.60 (1-2) per user-hour, relying on clusters of H100 GPUs distributed across the US and EU. While this may seem steep for streaming video, it pales in comparison to the costs associated with traditional film or game production, indicating a path forward for more accessible interactive experiences as efficiencies improve over time.
## The Future of Storytelling: A New Medium Emerges
As history has shown, each leap in technology heralds the evolution of storytelling—from ancient cave paintings to books, photography, radio, film, and video games. Odyssey posits that AI-generated \1 represents the next chapter in this rich narrative. If their vision materializes, we may be on the cusp of a transformative wave that reshapes entertainment, education, advertising, and beyond. Imagine training modules where learners can practice skills in a simulated environment, or virtual travel experiences that allow users to explore global destinations from the comfort of their homes.
The current research preview may merely scratch the surface of this ambitious vision, serving more as a proof of concept than a fully-fledged product. However, it undeniably offers a tantalizing glimpse into the potential of \1, promising a future where AI-generated worlds serve as dynamic playgrounds rather than passive viewing experiences.
### Conclusion
In summary, Odyssey’s pioneering work in \1 technology not only sets a new standard for AI innovation but also invites us to consider the limitless possibilities that lie ahead. As they continue to refine their model and overcome existing challenges, we stand on the brink of an exciting new era in storytelling and interaction. For those eager to witness this evolution firsthand, the research preview is now available for exploration.
For further insights into the world of AI and big data, consider attending the AI & Big Data Expo, which will be hosted in Amsterdam, California, and London, featuring a wealth of industry leaders and thought-provoking discussions.