• Research interests:

  • Language model merging (model parameter merging)

  • My personal research interests (not currently doing - just interests):

    1. A hybrid model that adaptively switches between System-1 and System-2 modes.
    2. Test-time learning.

    ⇒ In the pursuit of human-like LLMs (though I remain somewhat skeptical of whether such a direction is necessary), humans engage in System-2 reasoning when confronted with out-of-distribution problems. Crucially, the outcomes of such deliberate, reasoning processes are not temporary with repetition and learning; they gradually become System-1 responses. From this perspective, both 1 and 2 are key to building more human-like AI.

• If you have any questions or thoughts after reading my paper reviews (red text reflects my thinking), feel free to contact me at jcy9911@kaist.ac.kr (for an open discussion!)

Papers