#Woosung
LightNews LightNews
grandad420.bsky.social
Kim Woosung says Hello Darlin' ❤️
ALT: Kim Woosung says Hello Darlin' ❤️
static.klipy.com
October 31, 2026 at 2:08 PM
grandad420.bsky.social
Kim Woosung says Hello Darlin' ❤️
ALT: Kim Woosung says Hello Darlin' ❤️
static.klipy.com
October 29, 2026 at 12:59 AM
antara-poster.bsky.social
WN Korsel divonis delapan bulan terkait rudapaksa di Gili Trawangan - https://www.antaranews.com/berita/5817471/wn-korsel-divonis-delapan-bulan-terkait-rudapaksa-di-gili-trawangan
WN Korsel divonis delapan bulan terkait rudapaksa di Gili Trawangan
Majelis hakim Pengadilan Negeri Mataram, Nusa Tenggara Barat, menjatuhkan pidana penjara delapan bulan kepada warga negara Korea Selatan Woosung Kim dalam ...
www.antaranews.com
October 28, 2026 at 11:58 AM
antara-poster.bsky.social
WN Korsel divonis delapan bulan terkait rudapaksa di Gili Trawangan - https://www.antaranews.com/berita/5817471/wn-korsel-divonis-delapan-bulan-terkait-rudapaksa-di-gili-trawangan
WN Korsel divonis delapan bulan terkait rudapaksa di Gili Trawangan
Majelis hakim Pengadilan Negeri Mataram, Nusa Tenggara Barat, menjatuhkan pidana penjara delapan bulan kepada warga negara Korea Selatan Woosung Kim dalam ...
www.antaranews.com
October 28, 2026 at 11:58 AM
jorge12gm.bsky.social
En esta cuenta se banca a Agust D - SUGA.

youtu.be/zCNqdvR9e_w?...
BTS (방탄소년단) Agust D Ft. Ryuichi Sakamoto & WOOSUNG 'Snooze' MV
YouTube video by elisssa
youtu.be
October 21, 2026 at 8:25 AM
randomhourlybook.bsky.social
📖 New Testament history
✍️ George Wöosung Wade
🗓 Published: 1922

🔗 https://openlibrary.org/works/OL16905086W
Cover of New Testament history by George Wöosung Wade
October 15, 2026 at 5:00 AM
nubreed.bsky.social
WOOSUNG THE MAN THAT YOU ARE
October 11, 2026 at 10:24 AM
cslg-bot.bsky.social
Woosung Koh, Hyunsoo Lee, Kyungjae Lee, Haeju Park, Dahyun Lee, Dasol Hwang, Moontae Lee: Continually Post-training Small Models via Synthetic Self-distillation https://arxiv.org/abs/2610.13113 https://arxiv.org/pdf/2610.13113 https://arxiv.org/html/2610.13113
October 12, 2026 at 6:43 AM
jons1977.bsky.social
BUSAN INTERNATIONAL FILM FESTIVAL 2026

.... STARS ARRIVES...
BIFF RED CARPET FASSHION👠
Shuhua x Roh Jeong-eui x Cho Yi-hyun x Jeon Do-yeon

www.youtube.com/watch?v=V1D5...

Busan welcomes stars as Hyunbin, Jung Woosung, Fan Bingbing ignite BIFF red carpet

PICKON / WOW MEDIA CHANNEL
Oct 6, 2026
부산국제영화제 레드카펫👠 드레스👗 모음 '베스트 드레서는 누구?' | 김다현x슈화x노정의x전소영x조이현x조여정x김혜자x전도연 [BIFF Red Carpet]
YouTube video by PICKCON / 픽콘
www.youtube.com
October 6, 2026 at 8:17 PM
nonbinarydetective.bsky.social
Of course... #loislane #woosung #tiimk
September 29, 2026 at 9:32 PM
paper.bsky.social
[30/30] 264 Upvotes, 3 Comments, 4 Posts, arXiv:2609.02737

🆕Language Models Can Control Their Own Attention

Namgyu Ho, Huzama Ahmad, Woosung Koh, Se-Young Yun, Tal Schuster, Cicero Nogueira dos Santos
Language models spend most of their attention on a small fraction of context, yet they read the entire KV cache to find the few tokens that matter. If the user asks about a previous detail in a 1M-token conversation, global attention layers must scan the full context to generate each token of the reply. A prominent approach mitigates this cost by pre-selecting relevant tokens via lightweight proxy scores, but this extrinsic scoring still incurs O(N) per step. We take an intrinsic approach motivated by the simple question: wouldn't the model already know which parts of the context are relevant? To this end, we introduce Declarative Attention (DA), a protocol that elicits the model to declare where it needs to attend within its chain-of-thought, partitioning generation into three modes: <global> (full context), <focus> (a specific region), and <local> (recent output only). The inference engine parses these declarations like tool calls and skips most of the KV cache read. Under zero-shot evaluation across 15 long-context tasks, DA on off-the-shelf models (Gemma-4-31B, Qwen-3.6-27B) significantly reduces total attended tokens during decoding (52.0%, 31.1%) with modest accuracy drops (1.27pp, 2.75pp) that shrink with model scale. DA unlocks a new axis of sparse attention, with further potential under training-based methods that future work can explore.
September 27, 2026 at 2:27 PM
giemma321.bsky.social
MY ANTHEM FOR WHEN CHUMP & HIS ILK ARE GONE!!
www.youtube.com/watch?v=so50...
WOOSUNG – Happy Alone (feat. B.I.) | Official Visualizer
YouTube video by WOOSUNG
www.youtube.com
September 21, 2026 at 9:27 PM
nanoatp.bsky.social
[30/30] 249 Upvotes, 3 Comments, 4 Posts, arXiv:2609.02737

🆕Language Models Can Control Their Own Attention

Namgyu Ho, Huzama Ahmad, Woosung Koh, Se-Young Yun, Tal Schuster, Cicero Nogueira dos Santos
Language models spend most of their attention on a small fraction of context, yet they read the entire KV cache to find the few tokens that matter. If the user asks about a previous detail in a 1M-token conversation, global attention layers must scan the full context to generate each token of the reply. A prominent approach mitigates this cost by pre-selecting relevant tokens via lightweight proxy scores, but this extrinsic scoring still incurs O(N) per step. We take an intrinsic approach motivated by the simple question: wouldn't the model already know which parts of the context are relevant? To this end, we introduce Declarative Attention (DA), a protocol that elicits the model to declare where it needs to attend within its chain-of-thought, partitioning generation into three modes: <global> (full context), <focus> (a specific region), and <local> (recent output only). The inference engine parses these declarations like tool calls and skips most of the KV cache read. Under zero-shot evaluation across 15 long-context tasks, DA on off-the-shelf models (Gemma-4-31B, Qwen-3.6-27B) significantly reduces total attended tokens during decoding (52.0%, 31.1%) with modest accuracy drops (1.27pp, 2.75pp) that shrink with model scale. DA unlocks a new axis of sparse attention, with further potential under training-based methods that future work can explore.
September 19, 2026 at 12:07 AM
mexicanotu.bsky.social
Woosung Choi 2026 NPC Worldwide Monsterzym Korean Natural Regional Overwhelm bodybuilding middleweight in navy blue posers
September 15, 2026 at 10:09 AM
mexicanotu.bsky.social
Woosung Choi 2026 NPC Worldwide Monsterzym Korean Natural Regional Overwhelm bodybuilding middleweight in navy blue posers
September 15, 2026 at 10:08 AM
katiegeeks.bsky.social
listening to woosung?
September 12, 2026 at 11:18 PM
katiegeeks.bsky.social
I figured this was the actual trigger... but woosung had a song called "phase me" that has me doing the same thing every time I hear it 😭
September 13, 2026 at 12:06 AM
beingnegativehurts.bsky.social
Here is the birthday countdown for Woosung the Leader of The Rose, who's birthday is 25th
September 11, 2026 at 2:39 PM
paper.bsky.social
[27/30] 306 Likes, 18 Comments, 4 Posts
2609.02737, cs․CL | cs․AI | cs․LG, 02 Sep 2026

🆕Language Models Can Control Their Own Attention

Namgyu Ho, Huzama Ahmad, Woosung Koh, Se-Young Yun, Tal Schuster, Cicero Nogueira dos Santos
Language models spend most of their attention on a small fraction of context, yet they read the entire KV cache to find the few tokens that matter. If the user asks about a previous detail in a 1M-token conversation, global attention layers must scan the full context to generate each token of the reply. A prominent approach mitigates this cost by pre-selecting relevant tokens via lightweight proxy scores, but this extrinsic scoring still incurs O(N) per step. We take an intrinsic approach motivated by the simple question: wouldn't the model already know which parts of the context are relevant? To this end, we introduce Declarative Attention (DA), a protocol that elicits the model to declare where it needs to attend within its chain-of-thought, partitioning generation into three modes: <global> (full context), <focus> (a specific region), and <local> (recent output only). The inference engine parses these declarations like tool calls and skips most of the KV cache read. Under zero-shot evaluation across 15 long-context tasks, DA on off-the-shelf models (Gemma-4-31B, Qwen-3.6-27B) significantly reduces total attended tokens during decoding (52.0%, 31.1%) with modest accuracy drops (1.27pp, 2.75pp) that shrink with model scale. DA unlocks a new axis of sparse attention, with further potential under training-based methods that future work can explore.
September 10, 2026 at 12:05 AM
ddaengbiased.bsky.social
I fully intended to hop over to Layover next, but the last 3 songs on D-Day are some of my favorite Bad Day songs so

tidal.com/track/290140...
Agust D, Ryuichi Sakamoto & WOOSUNG - Snooze (feat. Ryuichi Sakamoto, WOOSUNG of The Rose)
Listen to Snooze (feat. Ryuichi Sakamoto, WOOSUNG of The Rose) on TIDAL
tidal.com
September 8, 2026 at 4:35 PM
milocaao3.bsky.social
Friends, The Rose has a new album (great songs) and Woosung promised an MV if we got one song to reach 10M streamings on Spotify. So, if you could help Black Roses out and put CHEMICAL (chosen song for the goal) in your playlists it would be deeply appreciated. ❤️🌹

open.spotify.com/track/4xBNUM...
CHEMICAL
The Rose · ROSE · Song · 2026
open.spotify.com
September 4, 2026 at 12:31 PM
cscl-bot.bsky.social
Namgyu Ho, Huzama Ahmad, Woosung Koh, Se-Young Yun, Tal Schuster, Cicero Nogueira dos Santos: Language Models Can Control Their Own Attention https://arxiv.org/abs/2609.02737 https://arxiv.org/pdf/2609.02737 https://arxiv.org/html/2609.02737
September 3, 2026 at 6:40 AM
cssd-bot.bsky.social
Roser Batlle-Roca, Woosung Choi, Joan Serr\`a, Fabio Morreale, Wei-Hsiang Liao, Xavier Serra, Emilia G\'omez, Yuki Mitsufuji: On the Human and Computer Alignment of Attribute-Based Music Matches https://arxiv.org/abs/2609.00987 https://arxiv.org/pdf/2609.00987 https://arxiv.org/html/2609.00987
September 2, 2026 at 6:44 AM
lunarriviera.bsky.social
having said that 1) this album is almost all woosung which makes me narrow my eyes 2) none of the yearning harmony he and dojoon used to weave (cf. "insomnia") 3) just hajoon and taegyeom on the last track ("home") but i get goosebumps bc these are 100% dojoon's lyrics. and they're about breaking up
where's my home
August 31, 2026 at 8:04 PM
debatemx.bsky.social
The Rose y Black Rose sanan juntos: "Rosetopia" en México y la promesa de volver a encontrarse
Woosung, Dojoon, Hajoon y Taegyeom, integrantes de The Rose, tomarán un descanso indefinido. Pero antes, volvieron a cantar y sanar junto a sus fans mexicanos
dlvr.it
September 1, 2026 at 12:02 AM