PhD @MIT EECS | @ucberkeley CS, @berkeleyML
Berkeley, CA
- Tweets 43
- Following 253
- Followers 158
- Likes 477
Pinned Tweet
Visual deep dive on FlashAttention by hand ✍️
(drawn with Excalidraw)
winterrykim.github.io/blog/2…
4
37
297
12,150
Terry (Taehan) Kim retweeted
Linear attention is arguably the most naive RNN, but still massively outperforms traditional RNNs by maintaining a matrix state. So... what if we use a (triadic) outer product of three vectors and maintain a three-dimensional state?
Introducing: Triadic Linear Attention 🧵
31
104
854
61,151
Terry (Taehan) Kim retweeted
There’s lots of buzz around agentic harnesses for robots, particularly for long horizon tasks that require complex reasoning and memory. But what will it really take to turn a reasoning agent into a reactive, reliable and low-latency robot policy?
In our new paper, workspace models, we design a new memory architecture that acts as a latent harness for stronger reasoning models. Here’s why we think this might be the way forward:
(1/9)
8
55
202
29,662
I’m excited to share 3 connected milestones toward a more trustworthy and scalable design loop for integrated photonics: 1 paper published in APL Engineering Physics and 2 accepted for oral presentation at IEEE Photonics Conference 2026.
🧵 1/7
3
2
16
528
6/7 Together, these works form an early design loop:
reuse prior knowledge → identify fabrication-critical features → refine with physics-based verification
Our goal is an end-to-end, scalable, and evidence-driven design loop for photonic integrated circuits.
1
1
56
7/7 Grateful to @punhojark, my co-first author across all 3 works, and to Mohammad Ali, SooHyuk Cho, and Prof. Di Liang for these collaborations.
APL paper: doi.org/10.1063/5.0327107
#IntegratedPhotonics #InverseDesign #DataCenters #OpticalIO
1
99
Visual deep dive on FlashAttention by hand ✍️
(drawn with Excalidraw)
winterrykim.github.io/blog/2…
4
37
297
12,150
Next post: From Memory to Photonics
winterrykim.github.io/blog/2…
166
Everyone is talking about memory lately: Micron, SanDisk, etc.
Here, we zoom out from FlashAttention/device memory to the next bottleneck: data-center communication.
That is where photonics matters.
winterrykim.github.io/blog/2…
w/@punhojark
If interested, come see our work at ICML.
1
4
20
1,510
For context, this builds on my previous post on FlashAttention and device memory:
1
411
I had a fun time writing a deep dive on Diffusion Language Models - with an equation walkthrough and Excalidraw sketches ✏️
In Part 1, I focused on the method: what does “noise” even mean for text, and how do DLMs denoise back into tokens?
winterrykim.github.io/blog/2…
2
8
32
3,984
Vessl provides a great platform
Are you looking for GPU credits for your next research project? @vesslai will be in San Diego for @NeurIPSConf , so if you're interested in GPU credits, feel free to sign up for a meeting here: cal.com/lucas-nam/neurips-ve….
Hope to see you there!
#NeurIPS #NeurIPS2025 #GPU #VESSLAI
2
668
Heading to #NeurIPS in San Diego (AI4Science Workshop) to present our expanded RNA secondary structure dataset with structure-aware train–test splits for improved scalability and robustness.
Preprint:
tinyurl.com/nbaj8h3v
Dataset: zenodo.org/records/15319168
Come say hi!
2
377
Terry (Taehan) Kim retweeted
👋We’re excited to launch DSG2-mini, our newest AI protein design model, now available in DiffuseSandbox⏳🎁, our new app for protein binder design. Click through to design a protein yourself! diffusesandbox.com 1/
6
80
385
63,483
(4/5)
Especially want to thank the Innovative Genomics Institute for providing funding to travel and attend RECOMB 2025. And to the organizers @RECOMB_conf @IEEEembs.
1
275
I’m especially grateful to Prof. Jamie Cate and Dr. Conner Langeberg for their early guidance in helping me get involved in research. Looking forward to meeting other researchers and learning at RECOMB!
#RECOMB2025 #EMBC2025 #AI4Science #MLforBio
2
396