1. 探索可用于自动驾驶/机器人等业务场景世界模型的前沿4D重建与生成模型技术
2. 跟踪重建与生成技术领域最新的技术发展和研究成果，提出新的技术创新和改进方案
3. 探索可用于自动驾驶/机器人场景VLA的、有3D感知能力和物理世界常识VLM基座技术
4. 通过重写、扩充和生成等方法合成大规模、高质量的数据；如指令调优、偏好对齐、模型优化以提高LLM在各个阶段（预训练、SFT、RLHF）的能力
5. 开展大模型算法研究，包括但不限于：多模态LLM/VLM的预训练/监督微调/强化后训练、LoRA/P-Tuning等高效微调，模型量化、分布式部署算法，实现大模型算法研究与应用
6. 研究和实施稳健的评估方法，以评估大模型在各个阶段的表现，揭示其能力的潜在机制和来源，并利用这种理解来推动模型改进
7. 跟踪打磨领域最新的技术发展和研究成果，提出新的技术创新和改进方案
8. 参与以实习生为主体的大模型方向的开源研究体系，对外在学术界顶会发表研究成果Explore cutting-edge 4D reconstruction and generative model technologies for world models applicable to business scenarios such as autonomous driving and robotics.
Keep track of the latest technical advancements and research achievements in reconstruction and generative technologies, and propose innovative technical solutions and improvement plans.
Investigate foundational VLM technologies equipped with 3D perception and physical world commonsense for Vision-Language-Action (VLA) systems in autonomous driving and robotics scenarios.
Synthesize large-scale, high-quality datasets through rewriting, augmentation, generation and other approaches; conduct instruction tuning, preference alignment and model optimization to enhance the performance of LLMs across all stages (pre-training, SFT, RLHF).
Conduct research on large model algorithms, including but not limited to: pre-training/supervised fine-tuning/post-training via reinforcement learning for multimodal LLMs/VLMs, efficient fine-tuning methods such as LoRA and P-Tuning, model quantization, and distributed deployment algorithms, to realize research and industrial application of large model algorithms.
Research and implement robust evaluation methodologies to measure the performance of large models at all stages, uncover the underlying mechanisms and origins of their capabilities, and leverage such insights to drive model optimization.
Continuously follow and refine state-of-the-art technical progress and research outcomes in the field, and deliver novel technical innovations and improvement proposals.
Participate in the open-source research ecosystem for large models led by interns, and publish research findings at top academic conferences worldwide.
If you excel exceptionally in any of the above fields, please do not hesitate to reach out—this is still a rare match, and we would like to learn more about you. Thank you.
