The Kim Jaechul Graduate School of AI at KAIST

Joined March 2022
51 Photos and videos
๐Ÿ“ข Three incoming faculty members at KAIST AI, starting in August 2026โœจ Dr. Sehoon Kim from xAI (@sehoonkim418), Dr. Hyunwoo Kim (@hyunw_kim), and Dr. Seung Wook Kim (@seungkim0123), both from NVIDIA, will be joining KAIST AI as Assistant Professors Check their websites below๐Ÿงต
2
7
83
15,242
They plan to begin accepting motivated students starting from the Fall 2026 semester. If you're interested in joining their labs, please feel free to contact them! Sehoon Kim: sehoonkim.org/ Hyunwoo Kim: hyunw.kim/ Seung Wook Kim: seung-kim.github.io/seungkimโ€ฆ
1
1
777
KAIST AI retweeted
Can a robot understand the nonverbal signals you give in real time โ€” your pointing gestures, your gaze, the things you never put into words? Meet EDITH: a framework that lets robots comprehend and act on human nonverbal signals. project-edith.github.io ๐Ÿงต[1/n] @KAIST_AI #Robotics #HumanRobotInteraction #VLA #ProjectAria
1
19
64
7,678
KAIST AI retweeted
The reversal curse. Edits that don't suppress negations. Multi-hop updates that don't propagate. These look like separate bugs. Our ICML 2026 spotlight argues they may share a common geometric origin, visible only when you study how representations move under updates ๐Ÿงต (1/11)
3
17
80
8,933
๐Ÿ“ฃJoin us at the Global AI Frontiers Symposium 2026 in Seoul, right before ICML๐ŸŒ๐ŸŒŸ Featuring keynotes from Leslie Pack Kaelbling (@MIT_LISLab) and Noam Brown (@polynoamial), plus panel discussions with Kyunghyun Cho (@kchonyc) and Emily Black! aifrontiers.kr/
1
3
16
4,178
KAIST AI retweeted
Happy to share that our work, Reward Score Matching (RSM), has been selected as an Oral Presentation at the SPIGM Workshop at ICML 2026. RSM asks a simple question: The literature on RL fine-tuning for diffusion/flow models looks fragmented, but which differences are actually fundamental? ๐Ÿ”— arxiv.org/abs/2604.17415
2
12
82
6,682
KAIST AI retweeted
๐ŸšจMost AI agents solve only the problems users explicitly ask about. But what about the problems users havenโ€™t noticed yet? ๐ŸŒŠTIDE enables proactive multi-problem discovery, helping agents uncover hidden issues ๐Ÿ” and recommend actionable next steps โœ…. huggingface.co/papers/2606.0โ€ฆ
2
16
25
3,359
KAIST AI retweeted
Happy to share our #Interspeech2026 paper!๐Ÿ—ฃ๏ธ arxiv.org/abs/2509.17901 w/ @seo_minjoon @KAIST_AI #NAVERCloud Quite a few video-LLMs still process video muted. Auditing 10 benchmarks, we find heavy visual shortcuts. We then make listening practical by compressing audio tokens 25ร—
5
13
741
KAIST AI retweeted
Can MLLMs actually track what's happening in a video? Introducing VSTAT ๐ŸŽฏ, our new benchmark for visual state tracking. The tasks are simple: count cups, read typed words, count page flips. Humans solve them easily. MLLMs don't. vision-x-nyu.github.io/vstatโ€ฆ ๐Ÿงต [1/11]
10
66
235
159,338
KAIST AI retweeted
A new RLHF vulnerability identified ๐Ÿšจ RLHF can be exploited to optimize misaligned biases, such as ideological or promotional biases. We introduce Alignment Tampering, a vulnerability where the LLM undergoing alignment influences the preference dataset itself, causing RLHF to amplify undesired behaviors. ๐Ÿ’ป Paper & Code: alignment-tampering.github.iโ€ฆ #ICML2026 #AIAlignment @KAIST_AI, @MIT_CSAIL 1/N ๐Ÿงต
1
4
17
3,381
KAIST AI retweeted
What if your retriever could speak every language your data speaks? ๐ŸŒ Your answer might live in a document ๐Ÿ“„, a SQL table ๐Ÿ—ƒ๏ธ, an RDF knowledge graph ๐Ÿ”—, or a property graph ๐Ÿ•ธ๏ธ, and OmniRetrieval reaches into all of them, meeting each source in its own native query language instead of flattening everything into one lossy space. Paper: huggingface.co/papers/2605.2โ€ฆ
1
30
75
5,133
KAIST AI retweeted
Excited to introduce ๐Ÿง‘โ€๐ŸŽ“๐—Ÿ๐—ฒ๐—ฎ๐—ฟ๐—ป ๐—ณ๐—ฟ๐—ผ๐—บ ๐—ช๐—ฒ๐—ฎ๐—ธ๐—ป๐—ฒ๐˜€๐˜€๐—ฒ๐˜€ (LearnWeak)! A framework that automatically specializes small CUAs for specific domains by ๐ŸŽฏ๐˜๐—ฎ๐—ฟ๐—ด๐—ฒ๐˜๐—ถ๐—ป๐—ด ๐˜๐—ต๐—ฒ๐—ถ๐—ฟ ๐—ผ๐˜„๐—ป ๐—ณ๐—ฎ๐—ถ๐—น๐˜‚๐—ฟ๐—ฒ ๐—ฝ๐—ฎ๐˜๐˜๐—ฒ๐—ฟ๐—ป๐˜€ in data generation and training. ๐Ÿงต(1/7)
6
12
20
1,054
KAIST AI retweeted
๐Ÿš€ Releasing โœจAXPOโœจ an RL method to lift agentic reasoning models past their next scaling tier. Be it math, perception, or search, AXPO fixes the structural blind spot 'just add tools' recipes leave untouched. 8B beats 4x larger 32B baseline on Pass@4. from NVIDIA ๐Ÿงต (1/7)
4
40
188
14,455
KAIST AI retweeted
Introducing TRQAM! Internalizing a KL trust region inside the sampling SDE stabilizes off-policy RL fine-tuning of pretrained flow policies. With TRQAM, we lift offline RL success on 50 OGBench tasks from 46% to 68%. ๐Ÿงต [1/8] yonghdong.github.io/blog/trqโ€ฆ
8
21
50
2,779
KAIST AI retweeted
๐ŸšจNew Optimizer Paper AMUSE: Anytime MUon with Stable gradient Evaluation AMUSE combines Muon with Schedule-Free-style gradient evaluation for stable anytime training without LR decay. โ€ข Stronger 124M / 720M / 1B pretraining โ€ข Strong ImageNet / ViT fine-tuning performance.
16
40
322
43,298
KAIST AI retweeted
We are looking for talented people interested in AI for Science, including ML for molecules, materials, and scientific discovery. If you are interested, please feel free to DM or email me. I am happy to chat and answer any questions.
๐Ÿš€ KAIST AI is recruiting faculty members in Seoul!๐ŸŒ™ Planning to attend ICML? Join us there and help shape a brighter future of AI๐ŸŒŸ forms.gle/ER9DEaUYtja7yGes5
2
13
2,127
๐Ÿš€ KAIST AI is recruiting faculty members in Seoul!๐ŸŒ™ Planning to attend ICML? Join us there and help shape a brighter future of AI๐ŸŒŸ forms.gle/ER9DEaUYtja7yGes5
1
14
32
4,138
KAIST AI retweeted
๐Ÿ“ข New preprint out on contextual integrity (CI) and a new Product-of-Experts (PoE) view of self-distillation! Introducing SelfCI, a novel self-distillation framework that operationalizes CI by optimizing for the intersection of task utility and minimal disclosure. ๐Ÿงต๐Ÿ‘‡
1
11
30
3,573
KAIST AI retweeted
Our work shows that using reasoning models as evaluators improves evaluation quality with additional test-time compute, enabling stronger re-ranking of #lanugagemodel outputs & matching the gains of increased compute at generation time. Learn how: neclab.eu/research-groups/huโ€ฆ #NECLabs
2
4
306