[AI Research Div.] Research Scientist - World Action Models (경력무관 / 인턴)
Salary not listed • Seoul; Seoul, South Korea • Internship
Posted 5d ago
Remote Work Policy
On-site
Employment Type
Internship
Categories
AI Research Engineer
About the job
We are seeking a Research Scientist to join our AI Research division, focusing on World Action Models. This role involves pioneering new paths in game development through bold imagination and breakthrough technology to create unforgettable worlds for fans. Our AI Research division is dedicated to developing proprietary AI core technologies for both gaming and non-gaming sectors, enhancing existing business competitiveness, and supporting AI technology commercialization to create new value and innovation. We aim to implement intelligent agents that interact naturally with humans in both game and real-world environments by researching and developing next-generation AI core technologies like Multimodal Foundation Models and AI Agents. We also strive to create production innovation, new user experiences, and AI-based services and business opportunities by applying AI technologies to game development, services, and various other industries.
Responsibilities
- Develop MLLM from scratch, analyzing and reproducing major VLM architectures and learning recipes, and designing and executing a three-stage learning pipeline (visual alignment, pre-training, physical SFT) based on proprietary LLMs.
- Develop flow-matching based video/action generators.
- Research World Model pre-training, studying methods to learn general physical representations not dependent on domain by utilizing game engine data (PUBG, Enjoy, etc.) and real-world data, and comparing/validating pixel-space and latent-space representation methods.
- Perform task-aware post-training on completed models for specific tasks and verify actual performance through RL (e.g., ReinFlow) and simulation-based learning.
- Establish an evaluation system by designing a closed-loop evaluation framework and benchmarks to enable rapid iteration of model improvement cycles.
- Utilize large-scale learning infrastructure, designing and executing distributed training in a large-scale GPU cluster environment.
Requirements
- Ability to lead research and quickly understand machine learning/artificial intelligence papers to design experiments.
- Ability to adapt quickly to new domains.
- Effective communication skills for smooth work collaboration.