Decart hat Oasis vorgestellt, das als „erstes Echtzeit-AI-Weltmodell“ gilt. Es verwendet Frame-Vorhersage, um basierend auf großen Datenmengen Echtzeit-Gameplay zu generieren, jedoch ohne Objektpermanenz. Das System simuliert physikalische Regeln und Grafiken, wobei Benutzerinteraktionen durch Tastatur- und Mausinput erfasst werden. Derzeit produziert Oasis eine pixelige Version von Minecraft, und obwohl zukünftige Anwendungen vielversprechend sind, wird momentan nicht das Potenzial der Nutzerinteraktion ausgeschöpft. Stattdessen bleibt es weit hinter den Erwartungen zurück.
In recent years, AI has been exploring some intriguing avenues. Advocates have touted the potential of generative AI for texture creation, while some have successfully run Doom on a neural network. The anticipation for user prompt-driven games has been palpable, yet here we are, and today is not that day.
Decart, an AI company that appears to have emerged suddenly, has introduced Oasis, touted as ‚the world’s first real-time AI world model.‘
Oasis operates using frame prediction. The AI model is fed vast amounts of data, allowing it to predict the next frame. This is why it seems to generate blocks without retaining a memory of the environment—it lacks object permanence and doesn’t retain data about your surroundings over time.
According to Decart, ‚Oasis processes user input from the keyboard and mouse to create real-time gameplay while internally simulating physics, game mechanics, and visuals.‘
In essence, you’re not truly playing a game; rather, you are experiencing a simulation of one.
In its announcement for this innovative model, Decart hinted at its potential, stating:
‚Envision a scenario where this integration is so seamless that foundational models could enhance modern entertainment platforms by dynamically generating content tailored to user preferences. Alternatively, imagine a gaming experience that allows for new forms of interaction, utilizing textual and audio prompts to influence gameplay.‘
This indicates the possibility for users to craft their own experiences, increasing the stakes or altering game elements in real time. However, the current model falls short of this vision.
What Decart seems to be producing is a rudimentary version of Minecraft, and unfortunately, it looks quite pixelated. The end of the teaser for this new engine suggests, ‚Imagine what AI-generated experiences could look like if everyone had the ability to create them,‘ a task that Mojang has been managing effectively for nearly two decades.
If Decart’s claims hold true, AI-generated Minecraft could indeed be groundbreaking, but it’s crucial to recognize how far we are from the promised future. The challenge of recreating a well-established game differs significantly from the ability to imagine and produce unique experiences based on user input.
Decart aspires to reach a point where users can simply declare, ‚Picture a pink elephant chasing me,‘ and one will materialize. However, it seems to have difficulty replicating an outdated game, drawing from countless hours of Minecraft playtime, as reported by MIT Technology Review.
Additionally, there are questions of whether users actually desire this, the copyright issues surrounding data sourcing, and the implications of creating art that revolves solely around personal expression.
This technology was developed using Nvidia H100 cards, and despite Minecraft typically being an accessible game, this software is not. Its performance hovers around 360p at 20 fps. Future enhancements with Etched ASICs, Decart’s current partner, may allow it to run such technology at 4K resolution.
Not all applications of AI in gaming aim to simply replicate existing games; some genuinely benefit the average gamer. AI-driven Frame Generation can enhance performance on the same hardware, theoretically relying on generating future frames based on previously displayed information.
I attempted to try out Oasis by visiting the website, but after a significant wait, the platform failed to load. This doesn’t mean it’s inaccessible, but I couldn’t get in, and at this point, I’d prefer to just play the original version.