About
Hello, I am Jiwon Jeong, an AI Research Engineer at Lotte Innovate (AI Tech LAB, Language AI Team). I received my M.S. in Artificial Intelligence from Sungkyunkwan University, advised by Prof. Hogun Park.
My research centers on attention architecture and efficiency and mechanistic interpretability in large language models: how attention components are designed, what they compute, and how they shape model behavior across languages. My earlier work covered logical fallacy detection and commonsense question answering.
I am actively seeking a Ph.D. position in NLP/LLM, to study how language models work internally and to make them more efficient and reliable.
I also contribute to open source, with merged patches to Hugging Face Transformers.
News
Show moreShow less
Education
Sungkyunkwan University, Dept. of Artificial Intelligence 2022 β 2024
M.S. in Artificial Intelligence. Advisor: Prof. Hogun Park.
Research on Commonsense QA and Logical Fallacy Detection.
Kookmin University, Dept. of Electronic Engineering 2016 β 2022
B.S. in Electronic Engineering. Major in Electronic Systems.
Publications
2026 Β· Under Review
How Vision Transformers Represent Position: A Mechanistic Comparison of Absolute and Rotary Positional Encodings
Mahmoud Mannes, Nikhil Maturi, Parva Mehta, Jiwon Jeong, Aravind Singaram Kannappan
arXiv preprint arXiv:2610.13992
Title withheld β under double-blind review
ICLR 2027 (main conference)
Title withheld β under double-blind review
NeurIPS 2026 Workshop on Unifying Representations in Neural Models (UniReps)
2026
Beyond Query Linearity: The Case for Nonlinear Queries
Jiwon Jeong, Dun Li Chan, Andrew Ceniccola, Rahul Sharma, Simon Sang, Parva Mehta, Pietro Di Nicco, Karim Saraipour, Marko Karbevski
NeurIPS 2026 Workshop on SLMs for Agentic Systems (SLM-Agents), Poster
Beyond Recovery Rate: Post-Recovery Verification and Reporting Gaps in Failure-Aware Evaluation of LLM Agents
(볡ꡬμ¨μ λμ΄: LLM κΈ°λ° μμ΄μ νΈ μ€ν¨ λμ νκ°μμμ μ¬ν κ²μ¦κ³Ό λ³΄κ³ νλͺ© λΆμ)
Jiwon Jeong, Hyunmook Cha
The 7th Korea Artificial Intelligence Conference (μ 7ν νκ΅μΈκ³΅μ§λ₯νμ λν), Jeju, SepβOct 2026
2025
Large Language Models Are Better Logical Fallacy Reasoners with Counterargument, Explanation, and Goal-Aware Prompt Formulation
Jiwon Jeong, Hyeju Jang, Hogun Park
NAACL 2025 Findings
2022
Improving Commonsense-based QA Model through a Cycle-Encoder
Jiwon Jeong, Soyoung Lee, Hogun Park
KSC 2022 (Outstanding Paper Award, Top 10%)
Professional Experience
Lotte Innovate β AI Research Engineer Sep 2025 β Present
AI Research Engineer in AI Tech LAB, Language AI Team.
Teaching & Mentoring
LIKELION β Online Mentor (AI NLP Engineer Intensive Course, 3rd Cohort) Feb 2026
Mentoring 4 teams on AI-based service development projects. Code review, project coaching, and Q&A support.
Sungkyunkwan University β Teaching Assistant 2023 β 2024
TA for Language Model (SWE3032-41) and Deep Learning courses.
Activities
Pseudo-Lab (κ°μ§μ°κ΅¬μ) β Runner, "LLM Technical Report μ λ νκΈ°" (13th Cohort) Oct 2026 β Present
Runner in the "LLM Technical Report μ λ νκΈ°" (Reading LLM Technical Reports) study group at Pseudo-Lab. A 16-week online study that reads recent LLM technical reports in depth to understand model architecture, data, and training, and examines each report's claims against its evidence. The cohort plans to write a survey paper on LLM technical reports after the study.
Stationhouse β Research Contributor, ConspiracyBench Sep 2026 β Present
Contributing to Stationhouse, an open research community building infrastructure for open model research. Working on ConspiracyBench, an LLM evaluation benchmark β contributing to schema design and benchmark item submission.
Pseudo-Lab (κ°μ§μ°κ΅¬μ) β Runner, "Deep Dive into LLM" (12th Cohort) Mar 2026 β Jul 2026
Runner in the "Deep Dive into LLM" study group at Pseudo-Lab, a non-profit AI research community. Studied Stanford CS336: Language Modeling from Scratch (2026) through weekly online discussions, with plans to pursue a research project after the course.
B-Peach LAB (Tech for Impact) β AI Researcher Aug 2025 β Jan 2026
Designed evaluation metrics for an AI-powered text adaptation service for slow learners.
Selected Projects
Hugging Face Transformers β Open Source Contribution Mar 2026 β Present
[PR #45722] (merged Apr 2026) β Removed six dead dummy classes (BeamScorer, ConstrainedBeamSearchScorer, Constraint, ConstraintListState, DisjunctiveConstraint, PhrasalConstraint) left over after v5 cleanup. Fixed check_repo.py failure on PyTorch-less installs.
[PR #45486] (merged Apr 2026) β Reported and fixed a protobuf decode error masking bug in tokenizer loading.
[PR #44436] (merged Mar 2026) β Fixed a bug in continuous batching for multimodal models. Added input preprocessing and a fallback mechanism for non-text-only models.
MCP Server for ChatGPT Apps Jan 2026 β Present
Built Model Context Protocol servers for ChatGPT Apps (OpenAI). Designed middleware architecture connecting frontend, backend, and LLM-powered enterprise solutions.
Korean LLM Benchmark Dec 2025 β Jan 2026
Evaluation framework for Korean LLM performance across Korean NLP tasks with standardized metrics and reproducible pipelines. [GitHub]
Awards
KSC 2022 β Outstanding Paper Award Dec 2022
Top 10% of accepted papers.
Text Ethics Hackathon β 2nd Place Dec 2021
ELECTRA-based classification model for hate speech and biased text detection.