Hyunggi Chang

engineer @tenstorrent

[Hyunggi Chang @changh95]

@ducha_aiki @CVPR @pesarlin @Nik__V__ @jianyuan_wang I guess I’ll stay up all night again this year 😂

[Hyunggi Chang @changh95]

@ducha_aiki You can SSH into a server, then summon a single window tmux there. You can then later just tmux attach to resume.
Not the simplest approach, but definitely useable!

[Tenstorrent @tenstorrent]

That's a wrap on TT-Deploy!

ICYMI we launched Tenstorrent Galaxy™ Blackhole deployed at scale with industry winning benchmarks including:

- @deepseek_ai 671B at 350+ t/s/u, with a roadmap to 500 t/s/u at $6 per million tokens, extending the low-cost serving curve where GPU economics collapse
- 10x faster than real-time high-quality video gen  (720p, 81-frame video in 2.5 seconds) w/ @prodialabs
- 90% pass rate across thousands of random @huggingface models

Links to recap + recordings below ⬇️

[Daniel DeTone @ddetone]

Today we release Boxer, a new lightweight approach that lifts open-world 2D bounding boxes to *metric* 3D: https://t.co/5IZ0tPlqvr

Here we show Boxer in action on an egocentric sequence captured from smart glasses:

[Hyunggi Chang @changh95]

@elonmusk Next step should be more affordable Starlink in Korea. All existing telecom providers are compromised, and the only barrier to moving to Starlink is due to almost 10 times more expensive service. Drop the price and I’m moving everything to Starlink and Tesla!

[Hyunggi Chang @changh95]

@Atenov_D @k1rallik Wasn’t really Sigrid Jin though.
If you look at commit history, proper development is all done by yeachan heo

[Hyunggi Chang @changh95]

@AjdDavison Maybe OpenAI tries to make a ChatGPT scale geometric foundation model?
Or… ChatSLAM? 😄

[Jason Kim @jsonbot117]

Got some time this weekend, so I went for a mapping with my bro @changh95 to record the symbolic monuments. Almost got steamed to death by the weather though…

[koide3 @k_koide3]

LiDAR2Minecraft: The voxel pattern matching isn't working great yet, but it's already fun to watch.
ボクセル種類割当がまだあまり良くできてないけれど既に結構楽しい。

[Ryohei Sasaki@engineer @rsasaki0109]

Global-LVBA
Global LiDAR-Visual Bundle Adjustment

Global-LVBA is a globally consistent LiDAR–Visual Bundle Adjustment system for refinement after LiDAR-inertial-visual odometry (e.g., FAST-LIVO2).
If you want to push accuracy to the next level, Global-LVBA further optimizes:
point-cloud map consistency (reducing layering/stratification artifacts).

camera poses towards pixel-level reprojection accuracy.

This figure shows qualitative comparisons before and after Global-LVBA refinement from both texture and geometric-structure perspectives. The initial poses are provided by FAST-LIVO2.

[Hyunggi Chang @changh95]

@AjdDavison Super interesting!

[Hyunggi Chang @changh95]

@marty188586 dmake? Interesting 👀

[Hyunggi Chang @changh95]

@marty188586 How?! 👀

[Hyunggi Chang @changh95]

HMG Atlas reminds me of this

[Tenstorrent @tenstorrent]

Today at @CES 2026, we unveiled our first-gen compact AI accelerator device designed with @Razer. This small device with serious compute plugs-in to any laptop, making AI accessible to more devs.

Find it on display at CES all week: Level 4 of the Venetian, at Zeno 4702.

Learn more --> https://t.co/zi0fBoGqQP

[Hyunggi Chang @changh95]

I’m attending CES 2026.

If you’re interested in AI accelerators, high-performance RISC-V CPUs, robotics, or autonomous driving,

I’d be happy to connect and exchange ideas.

Feel free to reach out or send me a message.

[Hyunggi Chang @changh95]

@damnang2 @Semicon_player 일단 슬램 인사드립니다

[Jason Kim @jsonbot117]

I’ve been quite busy lately developing the Humanoid autonomy stack as a member of URobotics. Now after few months of hard work, our team is excited to share a short video showcasing the Humanoid autonomy stack, H-EDK.

This is a short footage of our field locomotion demo.

[Hyunggi Chang @changh95]

@AjdDavison @eric_brachmann Super interesting!