Renee Jia

AI researcher at Meta  |  ex-Google

Learning, Memory & Reasoning

Researching how intelligent models learn from sequential experience, maintain and update knowledge over time, and use internal information for reliable reasoning.

Renee Jia

Research Interest

I study how models learn and remember, mostly through the lens of mid & post-training.

Learning from sequential experience

Supervised fine-tuning and what its data actually teaches, RL and reward design for multi-turn agents, continual learning without forgetting.

Writing: From Memory to Learning·Every Sequence Model Is a Memory Algorithm

Maintaining and updating knowledge over time

Memory architectures for long-horizon agents, what post-training overwrites, plasticity versus retention, moving knowledge from context into weights.

Writing: Distillation in LLMs

Reasoning with internal information

Chain-of-thought faithfulness, shortcuts and reward hacking, metadata that silently steers behavior, evaluations that catch regressions.

Writing: From Decorative to Load-Bearing·What If the Model Has Already Made Up Its Mind?

The same questions show up in production as user sequence modeling and personalization at scale. There the history is long, the user is a real person, and the system has far less patience than a research question does. I like working on both ends of that gap.

My path has run from research at Georgia Tech, Waterloo, and Harvard on reinforcement learning in text adventure games, to user sequence representation, teacher modeling, and distillation at Google, and now long-horizon user memory and sequence representation learning at Meta.

Publications

2026

  • TMLR 2026
    From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought Renee Jia, Di Mu Transactions on Machine Learning Research (TMLR), 09/2026 code·dataset
  • NeurIPS 2026 accepted
    Correct Labels Are Not Enough: Reliable Metadata Can Silently Control Model Behavior Renee Jia, Jinzhou Wu NeurIPS 2026 Workshop: Trustworthy AI for Good (AI4Good)
  • AAAI 2026 under review
    Where Can a Language Model Change a Trade? A Decision-Transmission Audit of a Hybrid Quantitative Trading System Renee Jia, Di Mu, Y. Yu, T. Liu AAAI Conference on Artificial Intelligence (AAAI) 2026

2023

2022

Reviewer: NeurIPS, ACL, EMNLP, EACL, NAACL, CoNLL.

Elsewhere

Off the clock: teaching snowboarding as a PSIA-AASI Level 1 instructor, and playing tournament poker (results).

Contact