Quake II runs on Whamm, Microsoft's experimental AI for real-time gaming

WHAMM, an AI model by Microsoft, runs Quake II with low frame rates.

: Microsoft's WHAMM model runs Quake II, showcasing its generative AI capabilities despite technical challenges. WHAMM's parallel image token generation reduces latency compared to its predecessor, WHAM-1.6B. It uses a single week of curated gameplay data for training, highlighting its potential. While engaging, the model suffers from graphical and operational limitations, indicating it's an experiment rather than a complete solution.

Microsoft has made a groundbreaking advance in real-time gaming with its experimental AI model, WHAMM, which successfully runs Quake II. Designed as a generative AI model, WHAMM specifically utilizes a MaskGIT-style architecture for image token generation. This technique marks a significant departure from previous autoregressive methods, bringing about a reduction in latency. As a result, it promises smoother interactions during gameplay. However, it's important to consider that this model is an ongoing project designed to highlight potential applications rather than serve as a finished gaming solution.

WHAMM, building off its predecessor WHAM-1.6B, integrates faster visual output processes meant to enhance real-time gaming experiences. The model was quick to adopt significant design changes, allowing it to leverage data from just one week of curated Quake II gameplay, specifically focusing on a singular level. This stands in stark contrast to OHAM-1.6B’s considerable reliance on seven years of gameplay data. Notably, WHAMM achieves a higher visual resolution of 640 x 360 pixels, improving the image quality in its gaming output.

Despite these advancements, the model has a series of recognized shortcomings which highlight its current limitations. Users face graphical anomalies and delayed responses, detracting heavily from attempts at a fully immersive experience. In practical terms, while players can perform a range of basic actions such as shooting, jumping, and crouching, they face notable issues with consistency and accuracy in combat mechanics. A key example includes inconsistency in health-tracking and damage statistics.

Furthermore, WHAMM's operational constraints extend to its memory capabilities. It struggles with remembering objects exiting the player's field of view beyond nine-tenths of a second. This leads to odd player experiences like unexpected teleportation and spontaneous enemy appearances with changes in camera angles. The current setup restricts the model to just one level, further emphasizing that the experiment is far from ready to fully replicate the Quake II experience or a broader spectrum of games.

Microsoft's endeavor with WHAMM brings forth discussions around AI's potential to act as a complement to, rather than a replacement for, human creativity in the game development process. The Redmond-based company envisions future models addressing present deficiencies to eventually aid developers in crafting better, more immersive narratives using AI-augmented tools. Ultimately, WHAMM's development tech demo serves as a stepping stone, inviting further exploration into AI's expanding role within interactive media industries, such as gaming.

Sources: TechSpot, Microsoft Blogs, GamingTechDaily, AI Weekly Report, The AI Journal

gamingAIGenerative AI