NVIDIA's research team has officially made a groundbreaking move by open-sourcing and releasing a new foundational model named NitroGen. This innovative model has been meticulously trained with a singular focus: to excel at playing games. It boasts an impressive repertoire of over 1,000 game titles, ranging from blockbusters like The Witcher 3 and Cyberpunk 2077 to a diverse array of other genres. The model's versatility shines through as it effortlessly navigates RPGs, racing games, battle royale games, and both 2D and 3D gaming environments.

NitroGen operates by taking game video frames as its input and generating controller operation signals as its output. This seamless integration makes it compatible with virtually all games that support controllers, offering a universal gaming solution. One of NitroGen's standout features is its support for post-training, which allows it to swiftly adapt to new games with minimal fine-tuning or lightweight adaptation efforts. This demonstrates its remarkable potential for cross-game generalization, a feat that has long eluded AI gaming models.

To achieve such remarkable results, the NVIDIA research team constructed a large-scale dataset using publicly available game videos. They then employed large-scale behavioral cloning techniques for training, ensuring that NitroGen could learn and mimic human-like gaming behaviors effectively. Experiments conducted thus far have consistently shown that NitroGen performs exceptionally well across a wide range of gaming scenarios. Moreover, it can seamlessly transfer its acquired skills to new games, making it a true game-changer in the world of AI gaming.