Ralphthon @ICML "Auto Research" supported by Codex
Ralphthon ICML Auto Research skills: team-attention/ralphthon-icml
VESSL credits for approved guests: claim-vessl-credits.team-attention.com
Morning session deck, Korean + English: ralphthon-icml-presentation.team-attention.com/slides.bi.html
Following SF, Singapore, and Busan, this 5th edition celebrates ICML. This time, the theme is Auto Research. One track creates papers, and the other builds Review Agents that review them.
SF, Singapore, Busan에 이어 ICML을 기념해 열리는 5번째 에디션. 이번에는 Auto Research입니다. 한쪽에서는 페이퍼를 생성하고, 다른 한쪽에서는 페이퍼를 리뷰하는 Review Agent를 만듭니다.
Watch previous Ralphthon video
want to see more side events?
Road to ICML 2026 Seoul: Community Events
About Event (English)
Once the Ralph Loop starts, you cannot touch your coding agent directly. If you want to touch your laptop, you have to wear the lobster costume first.
After SF, Singapore, and Busan, this 5th edition celebrates ICML week in Seoul.
This time, the theme is Auto Research.
There are two tracks:
Track 1: AI Scientist — Build an agent that creates a paper-style research artifact.
Track 2: Review Agent — Build an agent that reviews the papers from Track 1 in an ICML-style review format.
The core experiment:
Can AI agents write papers, review papers, and help produce better research loops?
The papers made by AI Scientist agents will be reviewed by Review Agent agents. Judges will look at the papers, the reviews, and the agent workflows behind them.
Prizes
Track 1 Prize: $10,000 API credits + 6 months of ChatGPT Pro
Track 2 Prize: $10,000 API credits + 6 months of ChatGPT Pro
Reviewers(Judges)
Moontae Lee - Assistant Professor in Information and Decision Sciences at UIC
Kyungwoo Song - Assistant Professor @Yonsei University
Younghoon Kim — Planar, ex-OpenAI Trust & Safety
Kyoung Whan Choe, PhD — Research Engineer at RLWRLD, robotics and cognitive neuroscience
Jae Gon Kim - Xenoscube, Lead Member of Technical Staff
Sigrid Jin - Founder of instruct.kr
Jungyoon Lim - Math Inc, Founding Engineer
Daniel Nam - NeuroFusion, Researcher
Jonghwa Park — Founder of BELL Therapeutics, ex-KIST Researcher
Dasol Choi - AI Safety Researcher @AIM Intelligence
Hyun-jun Choi - PhD Candidate at Seoul National University
Hoseung Choi, PhD - Research Engineer at KC ML2
Previous Ralphthon Editions:
Ralphthon @SF supported by OpenAI: https://luma.com/kxoq82yq
Ralphthon @Singapore supported by OpenAI: https://luma.com/4hx7p0vs
Time Table
09:30 Registration, doors open, and team formation
10:00-11:00
Opening and Codex Goal
W&B for Auto Research
VESSL AI GPU introduction
11:00-12:30 Research specification
12:30-15:30 Ralph Loop
15:30-16:30 Human editing and final submission
16:30 Submission snapshot for automatic matching
16:35-17:00 Track 1 self-review and Track 2 Review Agent runs
----
18:30-19:00 First-round judging and finalist selection
19:00-19:30 Laptop/table poster session
19:30-19:45 Final judge deliberation
19:45-20:05 Track 1 and Track 2 winner Oral presentations
20:05-20:20 Awards and closing - Special Prize DJI mic mini!
Sponsors
Presented Partner: OpenAI Codex https://openai.com/codex/
Observability Partner: https://wandb.ai/
GPU Compute Partner: https://vessl.ai/
Running Partner: https://dalpha.so/
Venue Partner: https://d2sf.naver.com/ko
Want to become a sponsor? Contact bong@team-attention.com, or connect with Goobong Jeong on LinkedIn.
랄프톤에 대해 (Korean)
랄프톤은 일반적인 해커톤이 아닙니다.
Ralph Loop가 시작되면 코딩 에이전트를 직접 만질 수 없습니다. 노트북을 만지고 싶다면 먼저 랩스터 코스튬을 입어야 합니다.
이번 주제는 Auto Research입니다.
두 개의 트랙
Track 1: AI Scientist — 논문 형태의 연구 결과물을 만드는 에이전트를 만듭니다.
Track 2: Review Agent — Track 1에서 만들어진 페이퍼를 ICML 리뷰 포맷으로 리뷰하는 에이전트를 만듭니다.
핵심 실험은 이것입니다.
AI 에이전트가 논문을 쓰고, 논문을 리뷰하고, 더 나은 연구 루프를 만들 수 있을까?
AI Scientist 에이전트가 만든 페이퍼를 Review Agent 에이전트가 실제로 리뷰합니다. 심사위원은 페이퍼, 리뷰, 그리고 그 뒤의 에이전트 워크플로우를 함께 봅니다.
Host: Team Attention
Team Attention runs Ralphthon, RalphWOD, clubs, and AI-native community experiments across Seoul, SF, Singapore, and Busan. We build formats where people learn new ways of working with agents by doing them in public.
Team Attention은 서울, SF, Singapore, Busan에서 Ralphthon, RalphWOD, 클럽, AI-native 커뮤니티 실험을 운영하는 팀입니다. 사람들이 에이전트와 함께 일하는 새로운 방식을 말로 배우는 것이 아니라, 직접 실행하면서 배우는 포맷을 만듭니다.
Founder: Goobong Jeong