Skip to content
View junkim100's full-sized avatar
👀
:)
👀
:)

Highlights

  • Pro

Organizations

@EdwinMichaelLab

Block or report junkim100

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
junkim100/README.md

Hi, I'm Dongjun Kim

Master's student at Korea University's NLP&AI Lab, advised by Dr. Heuiseok Lim. I work at the intersection of LLM evaluation and mechanistic interpretability to make models measurable, transparent, and trustworthy.

What I work on

  • LLM Evaluation
    • Ability decomposition and benchmark auditing (mixture-of-abilities, contamination checks, robustness sweeps)
    • Reproducible pipelines, unified metrics, longitudinal tracking, leaderboard design
    • Evaluation that correlates with user-perceived capability and downstream utility
  • Mechanistic Interpretability
    • Circuits/features via sparse autoencoders, probing, attribution, and targeted patching/ablations
    • Causal tracing and intervention studies to identify mechanisms behind reasoning and coding
    • Model-edit-aware analyses to understand when changes help or harm capabilities
  • AI Safety and Reliability
    • Auditing models for harmful behaviors and failure modes (e.g., deception, bias, adversarial vulnerability)
    • Continuous knowledge editing with retrieval for time-evolving domains (e.g., law)

Get in touch

If you're working on evaluation, interpretability, or AI safety, I'm happy to connect.

Pinned Loading

  1. JunGrad JunGrad Public

    A robust N-D autograd library with comprehensive operations, neural network layers, and optimizers.

    Jupyter Notebook

  2. Benchmark-Profiling Benchmark-Profiling Public

    Code for EMNLP 2025 paper - Benchmark Profiling: Mechanistic Diagnosis of LLM Benchmarks

    Python 5 1

  3. Unveiling-Coding-Regions-in-LLMs Unveiling-Coding-Regions-in-LLMs Public

    Python

  4. evalchemy evalchemy Public

    Forked from mlfoundations/evalchemy

    Automatic evals for LLMs

    HTML 3 1

  5. reasoning-steering reasoning-steering Public

    Forked from nrimsky/CAA

    Making models reason through CAA

    Jupyter Notebook

  6. Korean-Legal_Knowledge-Editing Korean-Legal_Knowledge-Editing Public

    Modification and Deletion of Knowledge in Korean Legal Domain | KT-Korea University Joint Research

    Jupyter Notebook