About the team On the TikTok USDS Search Team, you will have the opportunity to develop and apply cutting edge machine learning technologies in real-time large-scale systems, which serve billions of search requests every day. Via advanced NLP and multi-modal models, our projects impact and improve the search experience for hundreds of millions of users globally. We embrace a culture of self-direction, intellectual curiosity, openness, and problem-solving. The main job directions include: 1. Exploring and developing large-scale language models and optimizing enterprise applications to the extreme; 2. Data construction, instruction tuning, preference alignment, and model optimization; 3. Implementation of relevant applications, including content generation, summary etc.; 4. Collaborating with cross-functional teams to produce and apply new science to more responsibly develop and deploy large language models 5. In-depth research and exploration of more usage scenarios in future life. Responsibilities: - Conduct research and develop state-of-the-art algorithms in various stages of the development of LLM, including continued pretraining, SFT, RLHF; - Investigate and implement robust evaluation methodologies to assess model performance at various stages, unravel the underlying mechanisms and sources of their abilities, and utilize this understanding to drive model improvements. - Using inference stage techniques such as RAG, CoT, Prompt Engineering to improve the model output - Improve the performance of AI Search in the TikTok app to provide better search experience for users
Minimum qualifications: • Bachelor or advanced degree in computer science or a related technical discipline. • 5 years of related experience in one or more of the following areas: NLP, LLM, RL. • Proficient coding skills and strong algorithm & data structure foundation. Preferred qualifications: - Candidates with top-tier conference papers, including ICML, NeurIPS, ICLR, CVPR, ICRA, KDD etc., relevant internship experience or winners of ACM competitions are preferred; - Experience in using data-driven methods to enhance the capability of LLMs through various stages of the model development - Experience in RAG, Prompt Engineering or other inference time methods to enhance the performance of the system
Read Full Description