AI Challenge

1st AI Challenge

Problem

Design a question–answering (Q&A) system that generates answers grounded in medical clinical guidelines and research papers. Specifically, if you use Retrieval-Augmented Generation (RAG), how would you design the chunking, retrieval, and citation components? In what situations can RAG produce incorrect or misleading answers, despite using external documents? Propose evaluation metrics to reduce and detect hallucinations in large language models (LLMs).
다음 질문에 대해 논리적으로 서술하시오. 의료 가이드라인·논문을 근거로 답변하는 Q&A 시스템을 만든다. RAG(Retrieval-Augmented Generation)를 사용한다면 chunking / retrieval / citation을 각각 어떻게 설계할 것인가? RAG가 오히려 틀린 답을 생성할 수 있는 경우는 언제인가? LLM의 “환각(hallucination)”을 줄이기 위한 평가 지표를 제안하라.

Award Recipients

Winning Submissions

Woocheol Jang

Yunjeong Choi

Reference (Jinseok Lee)

2nd AI Challenge

Problem

You are training a large Transformer model on a single GPU. Inference fits comfortably in GPU memory, but training consistently runs out of memory (OOM). You decide to apply activation checkpointing. Why?

Award Recipients

Winning Submissions

Page updated

Google Sites

Report abuse