본문으로 건너뛰기
Ralphthon.

RPH-006 / 2026-07-12 / In build

Ralphthon @ICML: Auto Research

An ICML-week Ralphthon focused on AI Scientist, Review Agent, peer review, meta review, and Area Chair workflows.

Core from Luma

Protocol brief

  • Ralphthon @ICML is framed as Auto Research: an experiment in whether AI agents can write papers, review papers, and improve research loops.
  • Track 1 is AI Scientist, where teams build an agent that produces research output in paper form.
  • Track 2 is Review Agent, where teams build an agent that reviews Track 1 papers in ICML-style review format.

Hosted by

Host line

Luma lists Goobong Jeong, Yongrae Jo, and Joonyoung Park as hosts for the ICML-week Auto Research edition.

Goobong JeongYongrae JoJoonyoung Park

Run order

Schedule

  1. Research problem framing
  2. Agent setup
  3. Autonomous review loops
  4. Judge review and public-safe wrapup

Evidence

Proof

Auto Research trackReview Agent trackAutonomous review loopICML Seoul bridgesponsor-readable report
Open public event page

Luma record

Operational notes

Venue

  • NAVER D2SF Gangnam.
  • 18F, Samsung Fire & Marine Insurance Seocho Tower, 14 Seocho-daero 74-gil, Seocho-gu, Seoul.
  • The page positions the event as part of ICML Seoul week.

Tracks

  • Track 1: AI Scientist. Build an agent that produces research output in paper form.
  • Track 2: Review Agent. Build an agent that reviews papers from Track 1 in ICML-style review format.
  • AI Scientist submissions are reviewed by Review Agent teams, and judges evaluate papers, reviews, and workflows.

Prizes

  • Grand Prize: $10,000 in OpenAI API credits plus 6 months of ChatGPT Pro.
  • Track 1 Prize: $5,000 in OpenAI API credits plus 6 months of ChatGPT Pro.
  • Track 2 Prize: $5,000 in OpenAI API credits plus 6 months of ChatGPT Pro.

Judging Panel

  • Younghoon Kim, Planar, ex-OpenAI Trust & Safety.
  • Kyoung Whan Choe, PhD, Research Engineer at RLWRLD.
  • Kyoung Whan Choe's listed background includes robotics and cognitive neuroscience.

Sponsor

  • Presented Partner: OpenAI Codex.
  • The Korean description says this is not an ordinary hackathon but a research loop operated through AI agents.
  • The Luma page connects the format to Ralph Loop, no-touching-agent rules, and the lobster costume.

Previous Editions

  • The ICML page references prior Ralphthon SF and Ralphthon SG editions.
  • The SG event is cited as proof that the format can travel across AI Engineer side-event circuits.
  • The page describes Team Attention as the host team behind the Ralphthon format.