B.S. in Precision Instruments & Computer Science, Tsinghua University, 2011-2015
Work experience
03/2024-Now: Senior Research Scientist at Meta GenAI
Working on improving LLAMA-3 model on model reasoning and instruction following via SFT and RLHF
09/2020-02/2024: Senior Applied Scientist at Amazon AGI
Served as a tech lead on the ChatGPT-like Large Language Model project (Amazon Olympus Model) for model reasoning workstream
Devised new methods to align LLMs with human feedback, such as reward based data augmentation, constitutional critique and revision, RLAIF, self-alignment, etc.
Published a SOTA method for knowledge grounded response generation with web-search based question answering
Published SOTA models for enabling bots to address out-of-API user queries with external unstructured knowledge sources
Published a SOTA utterance rewriting model to resolve co-references and ellipsis to help bots better understand dialogue context
Launched better models for detecting and reducing contradiction and factual inconsistency in bots’ responses
Published new datasets and models for selecting better responses out of many candidates for open-domain chit-chatting