AGI Model Architect / Research Scientist in AGI Model Architecture

Responsibilities:
Design unified large model architectures with integrated capabilities in multimodal perception, reasoning, memory, and generation (across vision/audio/text).
Build systems that support continual learning, hierarchical memory, autonomous exploration, and self-evolution.
Advance the development of agent-based systems with autonomous task planning, cross-modal interaction, tool usage, and self-improvement capabilities.
Contribute deeply to the design of core components such as general representation learning, synchronized audio-visual modeling, world models, and sparse modeling.
Key Research Areas:
Multimodal Unified Architecture: Native co-frequency modeling and cross-modal reasoning across vision, speech, and language.
Continual Learning & Memory Mechanisms: Architectures that separate long-term memory from the core model to enable memory recall and task transfer.
World Modeling & Causal Reasoning: Enabling models to predict environmental states, plan behaviors, and update cognitive structures dynamically.
Sparse & Modular Architectures: Scalable, efficient, and interpretable ultra-large sparse model design.
Self-Evolution & Active Data Generation: Mechanisms for self-growth through reinforcement learning, self-supervision, and environment interaction.
Cross-Modal Understanding & Generation: Strengthening joint generation and decision-making capabilities in real-world physical environments.
Intelligent Agent Capability Transfer: Systematic enhancement of task generalization and tool-composition skills.
 
Work Location: US-Washington-Bellevue

Requirements:
Expertise in Transformer-based architectures and their applications in language and multimodal domains.
Hands-on experience in building or optimizing billion-scale models; familiar with training paradigms such as SFT (Supervised Fine-tuning), RLHF (Reinforcement Learning with Human Feedback), and self-supervised learning.
Preferred qualifications include deep understanding or practical experience in one or more of the following areas:
Multimodal models (e.g., vision-language models, audio-video models)
Reinforcement learning and autonomous agent systems
Complex reasoning and planning (e.g., search + LLMs, world modeling)
Sparse modeling and dynamic routing mechanisms
Strong engineering and system thinking capabilities, with the ability to translate cutting-edge research into production-level AGI model systems.
Publications in top-tier conferences/journals such as NeurIPS, ICLR, CVPR, ACL, etc., are highly desirable.。加分项：