ResearchMedium importance
Gamer, a vision-language-action model that plays in real time, tops the day's most upvoted papers on Hugging Face
The paper presents Gamer, a generalist agent for long tasks in interactive environments, which combines perception, memory and multi-step reasoning and works both on 2D interfaces and in simulated 3D worlds. The authors continued pre-training on tens of thousands of hours of real keyboard and mouse interaction, used instruction tuning with “sparse thinking” to cut redundant computation and trained the policy with multi-turn reinforcement learning. A cloud-edge inference engine runs many games at once with low latency. According to the paper, in several open-world 3D games Gamer achieves efficiency comparable to skilled players, as well as competitive results on 2D game benchmarks.
Published in edition no. 34 of Noticiário do Joãozinho ·
Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese
Original source
https://huggingface.co/papers/2611.18397