Joo-Kyung Kim (@supakjk) profile picture

Interested in conversational AI, mixed-initiative interaction, LLM post-training, LLM evaluation, multi-agents, and so on.

Kirkland, WA
Joined August 2009
Photo by Joo-Kyung Kim (@supakjk)Photo by Joo-Kyung Kim (@supakjk)
We are actively recruiting PhD research interns for 2026 at Amazon Alexa AI. We are particularly interested in candidates with experience and publication records in multi-turn/agentic reinforcement learning, or LLM with tool/skill/episodic memories. If you are interested in this opportunity, please email me at [email protected] with your CV and a brief statement of research interests.
We are recruiting PhD research interns for 2026. We focus on generative AI areas such as long-horizon reinforcement learning, LLM with tool/skill/episodic memories, LLMaaJ with non-verifiable rewards, and multi-modal agents, but not limited to these. The ideal candidates are current PhD students with 1st-author publications in top NLP/ML venues such as ACL, NAACL, EMNLP, NeurIPS, ICLR, and ICML. If you are interested in this opportunity, please email me (jookyk at amazon.com) with your CV. linkedin.com/jobs/view/43367…
2
10
150
17,935
We are recruiting PhD research interns for 2026. We focus on generative AI areas such as long-horizon reinforcement learning, LLM with tool/skill/episodic memories, LLMaaJ with non-verifiable rewards, and multi-modal agents, but not limited to these. The ideal candidates are current PhD students with 1st-author publications in top NLP/ML venues such as ACL, NAACL, EMNLP, NeurIPS, ICLR, and ICML. If you are interested in this opportunity, please email me (jookyk at amazon.com) with your CV. linkedin.com/jobs/view/43367…
10
30
1
315
47,527
왜 건물가치를 평균 임대료만 따질까요? 입주율과 임대료를 같이 고려해서 전체적인 수익으로 따져야하지 않을까요? 텅텅 빈 가운데 아주 일부만 높은 임대료로 들어가 있다면, 그걸 누가 가치있게 쳐줄까요?
전세계 대도시 공실 많은 지역 다 똑같음. 부동산 가치 안떨어지게 하려고 공실이라 월세 수익 없어도 그냥 둠. 부동산은 대표적으로 유한한 자원인데, 소수가 평생 반독점하면서 비상식적으로 운영하는데, 공실 3개월 이상되면 그냥 몰수해야함. (이러면 공사한답시고 비계 쳐놓고 공실 유지함.)
3
3,240
Joo-Kyung Kim retweeted
[ 케데헌에서 서울을 묘사할 때 특히 신경썼던 부분 ] - "마천루만 빼곡하면 뉴욕 느낌 주의!" - "초록이나 산 없이 건물만 빽빽하면 도쿄 느낌 주의!"
4
4,971
631
10,283
2,287,741
I'll be at #ACL2025NLP for presenting our MAPoRL paper (Hall 4/5, 10:30-12 on 7/29).
(1/x) Excited to share our new work on MAPoRL🍁: Multi-Agent Post-Co-Training for Collaborative LLMs with RL. Most current approaches just prompt pre-trained models and hope they’ll work together. But can we train LLM to discover the collaboration strategy?
8
930
(1/x) Excited to share our new work on MAPoRL🍁: Multi-Agent Post-Co-Training for Collaborative LLMs with RL. Most current approaches just prompt pre-trained models and hope they’ll work together. But can we train LLM to discover the collaboration strategy?
10
5
2
47
5,802
I am looking for emergency area chairs for ARR (commitment to EMNLP). The task is providing the meta review for a paper submission considering 3 existing reviews. (The topics are RAG or LLM bias). The hard due is tomorrow. Please let me know if you are interested in.
2
1
4
2,050
Adam deserves the award, but in Singapore everyone still uses SGD
21
61
17
783
110,206
"post-pre-training" is one of the weirdest terms I've ever seen while working in the field...
3
349
I am looking for an emergency reviewer for an ARR submission (ACL targeting). The paper is about post-training for Q/A and should be done by EoD. Please DM me if you are interested in the review!
2
1
1,815
I’ll be presenting my internship work about Chain-of-Instructions ⛓️ tomorrow in #AAAI2025 ! Come check it out if you are curious about LLM’s reasoning capability on compositional instructions! 🕛 12:30pm 📍Exhibition Hall E Poster Board 464
Instructions may compose of subtasks: the output from a subtask becomes the input for the next subtask, or called as “chain-of-instructions” (CoI)⛓️ CoI-tuned LLMs perform better on complex tasks & downstream tasks of multilingual summarization! 📎: arxiv.org/pdf/2402.11532.pdf
2
14
2,065
FSD가 잘된다 가정시 라이렌싱 하려면 카메라 위치 뿐 아니라 대부분 핵심기능들이 테슬라와 직접적으로 호환되야 하는데 그러면 그걸 채택하는 회사들은 그냥 껍데기, 인테리어, 조립 정도만 담당하는 위탁생산소로 전락할텐데.. 근데 그거라도 하는게 연명하는 방법일 수도...
236
사람들이 스푸트니크 쇼크를 얘기하는데 비슷한 부분도 있지만 좀 다른 것 같다. 그때는 미국이 투자 및 교육에 부족한 부분이 있었기 때문에 밀렸던 것 같은 반면 지금은 어마어마한 GPU비용 및 인건비를 쓰고 있던게 일시적으로 헛짓이 되었다보니. 오히려 헝그리정신이 더 필요한 타이밍이려나?
234
Joo-Kyung Kim retweeted
it be like that!
58
139
38
1,790
156,313
Joo-Kyung Kim retweeted
현 정부에서 이어졌던 이해할 수 없는 기괴한 일들이 정점에 이르렀습니다. 시간을 더 끌면 안될 비정상적인 상황입니다. 국회의 결단이 시급합니다. 세계가 주시하는 이 경악스러운 상황을 국회가 하루빨리 종식시키고 국민들에게 평온을 돌려줄 것을 촉구합니다.
865
35,987
3,449
36,472
3,636,448
Joo-Kyung Kim retweeted
추미애 의원의 과거 경고가 소름돋을 만큼 정확하다.ㅠ
20
6,454
109
6,409
365,212
Joo-Kyung Kim retweeted
외신에 이거 알려져서 나치즘 동경한다고 딱 한줄만 나가면 게임끝인데 기자님들....
[속보] 내란수괴범 김용현, 가장 즐겨 읽던 책 아돌프 히틀러의 <나의 투쟁>으로 밝혀져 씨발 진짜 할말이 없네
19,489
292
13,274
1,278,688