Jiale Zhao

jialeuuz@gmail.com jialeuuz OpenReview LinkedIn ORCID CV

Hi! I’m Jiale Zhao, a research assistant at Tsinghua University working on AI for Science. My current research combines genomics and large language models (LLMs) to study gene evolution. I earned a B.Eng. in Computer Science from Chongqing University of Posts and Telecommunications in 2026 and will begin a CS Ph.D. at Penn State University (PSU) in Spring 2027. I also work on co‑evolving documentation for coding agents (CoEvoDoc) and behavior‑oriented persona learning.

My research interests center on AI for Science and large language models, including:

  • AI for Science: genomics, LLMs, and gene evolution
  • Human-centered human–AI interaction (HCI)
  • Agent-based LLMs and multi-step reasoning
  • Rubric-based RLVR
  • Self-evolving systems
  • Interpretability and controllability
  • End-to-end multimodal interactive systems (e.g., GPT-4o)

Experience and Education

  • Sep 2026 – present: Research Assistant, Tsinghua University (AI for Science)
  • Spring 2027: Incoming Ph.D. student in Computer Science, Penn State University (PSU)
  • Sep 2023 – Jul 2026: LLM Algorithm Intern, Li Auto
  • Sep 2021 – Jun 2026: B.Eng., Computer Science, Chongqing University of Posts and Telecommunications

Selected Publications

For full links and the complete publication list, see the Publications page.

Accepted

When and What to Ask: AskBench and Rubric-Guided RLVR for LLM Clarification

Jiale Zhao, Ke Fang, Lu Cheng

Findings of ACL 2026 First author

RubricHub: A Comprehensive and Highly Discriminative Rubric Dataset via Automated Coarse-to-Fine Generation

Sunzhu Li, Jiale Zhao, Miteto Wei, Huimin Ren, Yang Zhou, Jingwen Yang, Shunyu Liu, Kaike Zhang, Wei Chen

ACL 2026 Main Second author

ThinkPilot: Steering Reasoning Models via Automated Think-prefixes Optimization

Sunzhu Li, Zhiyu Lin, Jiale Zhao, Shuling Yang, Chen Wei

Findings of EACL 2026 Co-author

Decoding the Ear: A Framework for Objectifying Expressiveness from Human Preference Through Efficient Alignment

Zhiyu Lin, Jingwen Yang, Jiale Zhao, Meng Liu, Sunzhu Li, Benyou Wang

Interspeech 2026 Oral Third author

Breaking the Exploration Bottleneck: Rubric-Scaffolded Reinforcement Learning for General LLM Reasoning

Yang Zhou, Sunzhu Li, Shunyu Liu, Wenkai Fang, Kongcheng Zhang, Jiale Zhao, Jingwen Yang, Yihe Zhou, Jianwei Lv, Tongya Zheng, Hengtong Lu, Chen Wei, Xie Yan, Mingli Song

ICML 2026 Main Sixth author

Under Review

Beyond Outcome Rubrics: Process Rewards for Non-Verifiable Tasks

Under Review

Not All Correct Responses Are Equal: Behavior Matters in Synthetic Data Generation for Post-Training

Under Review

Ongoing Work

  • CoEvoDoc — first-author collaboration with Prof. Lu Cheng (PSU) on a co‑evolving repository documentation system for coding agents that adapts structure complexity to project change magnitude.

Selected Projects

Projects from my Li Auto internship (Sep 2023 – Jul 2026).

Data Flywheel for Code LLM

Developed an evaluation-centric loop (SFT → evaluate → data build → filtering → back to SFT) to raise coding capabilities.

  • Evaluation-first: Standardized harnesses and rubrics to make harder coding tasks and underspecified requirements visible in evaluation.
  • Linked loops: Used evaluation feedback to guide harder data construction and filtering; filtered data re-entered the generation stack for repair and reuse.

Multi-step Reasoning + Tool Invocation Agent

Developed a code-LLM agent that plans, writes code, and executes tool calls for precise answers.

  • Multi-step reasoning: Broke complex or code-debugging tasks into structured plans so context could be stitched into a single executable query.
  • Tool grounding: Integrated function calls and code execution for real-time data, external APIs, and environment actions when model priors or knowledge bases fell short.

MindGPTo (GPT‑4o-style multimodal app)

Built an end-to-end audio + vision application with paralinguistic control and a modular FE/BE split.

  • Mode coverage: Supported audio→ASR→LLM→TTS, audio2text→TTS, end-to-end audio2audio, and multimodal audio+image+video→text→TTS workflows.
  • Paralinguistic SFT: Built large-scale audio data pipelines for colloquial speech and nuanced cues (beyond laughter/pauses) such as age, gender, compound emotions, emotional actions, and ambient sounds.