Yoonsu Kim

PhD student in @kixlab_kaist, SoC, KAIST. Interested in HAI, Human-AI symbiosis, Generative AI and Education, and tech with Good influence on the world🌏❤️!

[Yoonsu Kim @yoons__u]

When you're working with a GenAI tool — how do you actually communicate what you want?

Our #DIS2026 paper "IntentFlow" investigates how LLM interfaces can support the full process of intent communication 🧵

(1/6)

[Yoonsu Kim @yoons__u]

🌐 https://t.co/BxvXJDF8Rz
📄 arXiv: https://t.co/QJDdSoAfAB

Presenting at #DIS2026 🇸🇬
📅 Wed, 17 Jun | 9:48–10:00 AM
📍 LT 52, Stephen Riady Centre Lv2
w/ @kihoon_son, @seoy_kim, Brandon Chin, and @imjuhokim

#DIS2026 #HCI #LLM #GenerativeAI #HumanAIInteraction

[Yoonsu Kim @yoons__u]

When you're working with a GenAI tool — how do you actually communicate what you want?

Our #DIS2026 paper "IntentFlow" investigates how LLM interfaces can support the full process of intent communication 🧵

(1/6)

[Yoonsu Kim @yoons__u]

Design implications for LLM interfaces:

🔗 Bidirectional intent-output traceability
🔭 Synchronized views as scaffolds for exploration
📌 Progressive commitment — from loose exploration to pinned intents
🗂️ Visible intent spaces to contextualize new articulation

(5/6)

[Yoonsu Kim @yoons__u]

Key insight: intent communication is cyclic, not linear.

🌀Articulate → explore → stabilize → verify → and loop back to refine.

Supporting the full cycle reduced corrections and cognitive load — shifting users from reactive fixing to actively steering output.

(4/6)

[Yoonsu Kim @yoons__u]

We built IntentFlow — a research probe for LLM-based writing that supports all four:
🔹 Intent Panel with adjustable dimensions
🔹 Version history with rollback
🔹 Hover-based intent-to-output linking

(3/6)

[Yoonsu Kim @yoons__u]

From a review of 46 HCI papers, we identified 4 key aspects:

✏️ Articulation — vague goals → editable intents
🔭 Exploration — discovering new directions
📋 Management — persisting intents across turns
🔗 Synchronization — verifying output reflects intent

(2/6)

[DaEun Choi @daeun_choi_]

GenAI is good at producing a result.
But early-stage design is not about a single result —it is about exploring possibilities.

Our #DIS2026 paper “IdeaBlocks” won an Honorable Mention 🏆

We ask: how can designers express not only what to generate, but how to explore?

(1/n)

[Saelyne Yang @saelyne_yang]

🖥️ Most GUI agents focus on automation -- but what about understanding the *user*?

In our upcoming #CVPR2026 paper, we introduce GUIDE, a benchmark for understanding & assisting users in open-ended GUI tasks! 🎉

[Kihoon Son @kihoon_son]

When creative professionals share their workflow, hundreds of decisions lie behind it 👀

What if we could capture the reasoning, including the implicit reasoning ☁️, not just the how-to?

What if their AI agent 🤖 understood this and learned to decide the way they would?

➡️🧵

[Kihoon Son @kihoon_son]

AI agents are taking on more and more of our work (even creativity-related work 😱), but is handing off and reviewing the only way humans and agents can collaborate?

We think there's a better way. 🧵

#Agent #Figma #Design #HAI

[Tae Soo Kim @tae_skim]

AI agents are running complex workflows, and writing full codebases and docs. But how do you verify these outputs at scale?

An LLM judge gives you "3/5" with a broad reason. Which parts caused that score? Any patterns across outputs?

We built Evalet 🔬 to fix this. CHI'26

[Kihoon Son @kihoon_son]

Pleased to share our upcoming #CHI2026 workshop: Herding CATs: Making Sense of Creative Activity Traces. 📈🎨🐈‍⬛

We invite 2-4 page position papers on analyzing and understanding the creative process through digital traces.

🔗 https://t.co/MwwNZVCjzT📅 Due: Feb 12, 2026

[Kevin Pu @kevpjk]

I am presenting ProMemAssist at #UIST2025!
The demo is funny -- check out how my AI assistant on smart glasses helps me pack 🎒👓
📍Sept 30 Tue | 12:12 PM | Grand Ballroom

[Kihoon Son @kihoon_son]

Are you interested in capturing professionals' tacit knowledge—like implicit decisions—and reusing it in generative agents 🤖 to better support creative work 👨🏻‍🎨?

👉 I will present my thesis work at UIST25 Doctoral Symposium (Poster: Tue evening).

See you in Busan! 🇰🇷 #UIST2025

[Yoonseo Choi @yoon0u0]

Are you heading to Busan? 🏄‍♀️🌴🇰🇷
I'll present our paper 🪷BloomIntent: Automating Search Evaluation with LLM-Generated Fine-Grained User Intents at #UIST2025!

📍Oct 1 · 11:12–11:24 AM (Miami Room)

[Tae Soo Kim @tae_skim]

You ask ChatGPT to write a message for:
💼 Credit-stealing colleague → “Building on *my* idea…”
🏠 Messy roommate → “babe wake up new mold just dropped”

Same you. Same task. Different context.
Can LLMs learn this? 🤔

We built CUPID 🏹 to find out.
🔗 https://t.co/JrAdy2SbSV

[Yoonjoo Lee @yoonjoo_le2]

🎓I officially defended my PhD! Huge thanks to my amazing advisor @juhokim and committee @eytanadar @aliceoh @tongshuangwu @seo_minjoon.

This fall, I'm excited to join @UMichCSE as a postdoc with @QVeraLiao to continue my research in human-centered AI and cognitive alignment!💙

[Kevin Pu @kevpjk]

New paper at #UIST25!
🧠 Wearable AI assistants can support our daily tasks, but poorly timed interventions often hurt more than help.
🕶️ ProMemAssist is a wearable AI assistant that delivers proactive task support—at the right time—by modeling your working memory in real time.

[Yoonseo Choi @yoon0u0]

안녕하세요, @kixlab_kaist 에서 만든 📢 공약쏙쏙 시스템📢을 소개합니다! (https://t.co/HnAgGJ8tfR)

🤔공약쏙쏙?
#대선2025 공약을 바탕으로 공약이 가져올 수 있는 사회적 영향, 그리고 '나'에게 어떤 변화가 생�� 수 있을지 LLM (GPT-4o) 을 활용해 자동으로 보여주는 시스템입니다.

(1/3)

[Yoonsu Kim @yoons__u]

@sunniesuhyoung @orussakovsky @andresmh @jennwvaughan @QVeraLiao @parastooabtahi Big Congrats🎉🫶