Marco
@mcognetta.bsky.social
Language and keyboard stuff at Google + PhD student at Tokyo Institute of Technology.
I like computers and Korean and computers-and-Korean and high school CS education.
Georgia Tech → 연세대학교 → 東京工業大学.
https://theoreticallygoodwithcomputers.com/
I like computers and Korean and computers-and-Korean and high school CS education.
Georgia Tech → 연세대학교 → 東京工業大学.
https://theoreticallygoodwithcomputers.com/
Pinned
Marco
@mcognetta.bsky.social
· Jan 5
A lot of you followed me due to #NLP, but I like to post about #chess (especially computer chess), #programming (especially puzzles, code golf, etc), and machine learning.
And some less technical stuff like #Korean, #Esperanto, and #trains (mostly in Japan, just due to proximity).
And some less technical stuff like #Korean, #Esperanto, and #trains (mostly in Japan, just due to proximity).
Reposted by Marco
Alright, it's a tad rushed because it was written in between looking after the kids and I wanted to get it out before the new year but here it is. Reflections on 2025 and the biggest career change I've ever done.
samwho.dev/2025
samwho.dev/2025
2025
Reflecting on my path through 2025.
samwho.dev
December 30, 2025 at 5:17 PM
Alright, it's a tad rushed because it was written in between looking after the kids and I wanted to get it out before the new year but here it is. Reflections on 2025 and the biggest career change I've ever done.
samwho.dev/2025
samwho.dev/2025
Reposted by Marco
I didn't see enough news about this crazy/cool paper from DeepSeek - you take text, and convert it to imagery, and you get BETTER performance. Oh and some really neat ideas about contextual memory too.
www.distributedthoughts.org/a-picture-is...
www.distributedthoughts.org/a-picture-is...
A Picture Is Worth Ten Thousand Tokens
DeepSeek's new OCR model achieves 97% accuracy while using one-tenth the tokens by rendering text as images. The computationally "heavy" modality turns out to be the efficient one, which should make u...
www.distributedthoughts.org
December 29, 2025 at 7:27 PM
I didn't see enough news about this crazy/cool paper from DeepSeek - you take text, and convert it to imagery, and you get BETTER performance. Oh and some really neat ideas about contextual memory too.
www.distributedthoughts.org/a-picture-is...
www.distributedthoughts.org/a-picture-is...
Technical content will have a higher ratio on my profile soon™, but for now I'm just chillin.
I recently visited Jiufen (a seaside/mountain town east of Taipei; ostensibly the aesthetic inspiration for Spirited Away).
The highlight was the tea, but we also took a nice hike.
The highlight was the tea, but we also took a nice hike.
December 30, 2025 at 6:25 AM
Technical content will have a higher ratio on my profile soon™, but for now I'm just chillin.
I recently visited Jiufen (a seaside/mountain town east of Taipei; ostensibly the aesthetic inspiration for Spirited Away).
The highlight was the tea, but we also took a nice hike.
The highlight was the tea, but we also took a nice hike.
December 30, 2025 at 6:24 AM
I recently visited Jiufen (a seaside/mountain town east of Taipei; ostensibly the aesthetic inspiration for Spirited Away).
The highlight was the tea, but we also took a nice hike.
The highlight was the tea, but we also took a nice hike.
I also keep a list of terrible company names.
Sauron has dethroned my previous worst: Saltmine, a "workplace strategy, design & optimization software" company.
Sauron has dethroned my previous worst: Saltmine, a "workplace strategy, design & optimization software" company.
Palantir. Sauron. CyberDyne. Blackrock. Mr Beast Inc. Cerberus. Blackwater. Anduril. Citizen. Citadel. Gilead. Bain Capital. Darktrace. Oracle. Umbra.
December 29, 2025 at 9:25 AM
I also keep a list of terrible company names.
Sauron has dethroned my previous worst: Saltmine, a "workplace strategy, design & optimization software" company.
Sauron has dethroned my previous worst: Saltmine, a "workplace strategy, design & optimization software" company.
Reposted by Marco
Imagine you are a trained next word prediction model and you see this.
December 29, 2025 at 5:59 AM
Imagine you are a trained next word prediction model and you see this.
Reposted by Marco
The For You feed is going to be down while I'm upgrading the CPU from AMD 7900 (12 core) to AMD 9950X3D (16 core).
This should increase how many users it can serve by ~50%-100%.
If all goes well it should be back up in an hour or two.
This should increase how many users it can serve by ~50%-100%.
If all goes well it should be back up in an hour or two.
December 28, 2025 at 4:40 PM
The For You feed is going to be down while I'm upgrading the CPU from AMD 7900 (12 core) to AMD 9950X3D (16 core).
This should increase how many users it can serve by ~50%-100%.
If all goes well it should be back up in an hour or two.
This should increase how many users it can serve by ~50%-100%.
If all goes well it should be back up in an hour or two.
What is happening here?
December 28, 2025 at 1:34 PM
What is happening here?
Reposted by Marco
this X post is worded like an indictment of Bluesky, but it’s actually a perfect example of why Bluesky is *better* than X — users can’t create their own shit on X when they don’t like what X is doing
December 26, 2025 at 6:13 PM
this X post is worded like an indictment of Bluesky, but it’s actually a perfect example of why Bluesky is *better* than X — users can’t create their own shit on X when they don’t like what X is doing
Hong Kong: good public transportation, fast escalators, fun hills, tasty food.
But no Claude.
But no Claude.
December 27, 2025 at 4:25 AM
Hong Kong: good public transportation, fast escalators, fun hills, tasty food.
But no Claude.
But no Claude.
2026: 👀 is out, ꙮꙮ is in.
> Multiocular O (ꙮ) is a unique glyph variant found in a single 15th-century manuscript, in the Old Church Slavonic phrase серафими многоꙮчитїи (abbreviated мн̑оꙮ҆читїи̑; serafimi mnogoočitii, 'many-eyed seraphim').
en.wikipedia.org/wiki/Cyrilli...
en.wikipedia.org/wiki/Cyrilli...
December 26, 2025 at 12:14 PM
2026: 👀 is out, ꙮꙮ is in.
Reposted by Marco
Hi 👋 I’m Sam, I create visual essays about programming.
My posts from 2025:
- samwho.dev/reservoir-sa...
- samwho.dev/big-o/
- ngrok.com/blog/prompt-...
I also appeared on other folks’ websites:
- simonwillison.net/2025/Dec/19/...
- writethatblog.substack.com/p/sam-rose-o...
My posts from 2025:
- samwho.dev/reservoir-sa...
- samwho.dev/big-o/
- ngrok.com/blog/prompt-...
I also appeared on other folks’ websites:
- simonwillison.net/2025/Dec/19/...
- writethatblog.substack.com/p/sam-rose-o...
December 26, 2025 at 11:16 AM
Hi 👋 I’m Sam, I create visual essays about programming.
My posts from 2025:
- samwho.dev/reservoir-sa...
- samwho.dev/big-o/
- ngrok.com/blog/prompt-...
I also appeared on other folks’ websites:
- simonwillison.net/2025/Dec/19/...
- writethatblog.substack.com/p/sam-rose-o...
My posts from 2025:
- samwho.dev/reservoir-sa...
- samwho.dev/big-o/
- ngrok.com/blog/prompt-...
I also appeared on other folks’ websites:
- simonwillison.net/2025/Dec/19/...
- writethatblog.substack.com/p/sam-rose-o...
> Multiocular O (ꙮ) is a unique glyph variant found in a single 15th-century manuscript, in the Old Church Slavonic phrase серафими многоꙮчитїи (abbreviated мн̑оꙮ҆читїи̑; serafimi mnogoočitii, 'many-eyed seraphim').
en.wikipedia.org/wiki/Cyrilli...
en.wikipedia.org/wiki/Cyrilli...
December 24, 2025 at 5:35 PM
> Multiocular O (ꙮ) is a unique glyph variant found in a single 15th-century manuscript, in the Old Church Slavonic phrase серафими многоꙮчитїи (abbreviated мн̑оꙮ҆читїи̑; serafimi mnogoočitii, 'many-eyed seraphim').
en.wikipedia.org/wiki/Cyrilli...
en.wikipedia.org/wiki/Cyrilli...
Is this a neural network?
December 23, 2025 at 12:04 PM
Is this a neural network?
No wifi on my flight, so I spent some time clearing out my phone's chrome tabs.
I closed several hundred tabs, but still have the :D icon.
I closed several hundred tabs, but still have the :D icon.
December 23, 2025 at 9:17 AM
No wifi on my flight, so I spent some time clearing out my phone's chrome tabs.
I closed several hundred tabs, but still have the :D icon.
I closed several hundred tabs, but still have the :D icon.
I am convinced the GDP of Japan would rise significantly if the Shinkansen wifi was even moderately usable.
It is the perfect place to hack, other than that you can't access the internet at all >95% of the time.
It is the perfect place to hack, other than that you can't access the internet at all >95% of the time.
December 21, 2025 at 7:11 AM
I am convinced the GDP of Japan would rise significantly if the Shinkansen wifi was even moderately usable.
It is the perfect place to hack, other than that you can't access the internet at all >95% of the time.
It is the perfect place to hack, other than that you can't access the internet at all >95% of the time.
I'm looking to build a personal GPU cluster for bespoke model training/inference. Not LLMs, but things in the 50-100M parameter range.
I don't need absolute top-of-the-line, but something reasonably good.
I'm looking for recs, examples of existing setups, etc., any help is appreciated!
I don't need absolute top-of-the-line, but something reasonably good.
I'm looking for recs, examples of existing setups, etc., any help is appreciated!
December 21, 2025 at 3:48 AM
I'm looking to build a personal GPU cluster for bespoke model training/inference. Not LLMs, but things in the 50-100M parameter range.
I don't need absolute top-of-the-line, but something reasonably good.
I'm looking for recs, examples of existing setups, etc., any help is appreciated!
I don't need absolute top-of-the-line, but something reasonably good.
I'm looking for recs, examples of existing setups, etc., any help is appreciated!
Reposted by Marco
📍 Okayama, Japan 🇯🇵 — Local Time 09:00:11 PM
December 19, 2025 at 12:00 PM
📍 Okayama, Japan 🇯🇵 — Local Time 09:00:11 PM
Reposted by Marco
Aditya Yadavalli, Tiago Pimentel, Tamar I Regev, Ethan Wilcox, Alex Warstadt
What Do Prosody and Text Convey? Characterizing How Meaningful Information is Distributed Across Multiple Channels
https://arxiv.org/abs/2512.16832
What Do Prosody and Text Convey? Characterizing How Meaningful Information is Distributed Across Multiple Channels
https://arxiv.org/abs/2512.16832
December 19, 2025 at 5:24 AM
Aditya Yadavalli, Tiago Pimentel, Tamar I Regev, Ethan Wilcox, Alex Warstadt
What Do Prosody and Text Convey? Characterizing How Meaningful Information is Distributed Across Multiple Channels
https://arxiv.org/abs/2512.16832
What Do Prosody and Text Convey? Characterizing How Meaningful Information is Distributed Across Multiple Channels
https://arxiv.org/abs/2512.16832
Reposted by Marco
I want bicycle infrastructure that’s safe enough to do this
December 19, 2025 at 4:37 AM
I want bicycle infrastructure that’s safe enough to do this
FLaNN is one of the coolest communities I've been part of, and now they have their own workshop!
Enjoy some research (and submit your own) about interpretability, expressivity, etc. of neural networks through the lens of formal language theory!
Enjoy some research (and submit your own) about interpretability, expressivity, etc. of neural networks through the lens of formal language theory!
Announcing the first Workshop on Formal Languages and Neural Networks (FLaNN)!
We invite the submission of abstracts for posters that discuss the formal expressivity, computational properties, and learning behavior of neural network models, including large language models (LLMs).
We invite the submission of abstracts for posters that discuss the formal expressivity, computational properties, and learning behavior of neural network models, including large language models (LLMs).
December 19, 2025 at 3:51 AM
FLaNN is one of the coolest communities I've been part of, and now they have their own workshop!
Enjoy some research (and submit your own) about interpretability, expressivity, etc. of neural networks through the lens of formal language theory!
Enjoy some research (and submit your own) about interpretability, expressivity, etc. of neural networks through the lens of formal language theory!
Reposted by Marco
Announcing the first Workshop on Formal Languages and Neural Networks (FLaNN)!
We invite the submission of abstracts for posters that discuss the formal expressivity, computational properties, and learning behavior of neural network models, including large language models (LLMs).
We invite the submission of abstracts for posters that discuss the formal expressivity, computational properties, and learning behavior of neural network models, including large language models (LLMs).
December 19, 2025 at 2:59 AM
Announcing the first Workshop on Formal Languages and Neural Networks (FLaNN)!
We invite the submission of abstracts for posters that discuss the formal expressivity, computational properties, and learning behavior of neural network models, including large language models (LLMs).
We invite the submission of abstracts for posters that discuss the formal expressivity, computational properties, and learning behavior of neural network models, including large language models (LLMs).
Reposted by Marco
📍 Kyoto, Japan 🇯🇵 — Local Time 09:00:40 PM
December 18, 2025 at 12:00 PM
📍 Kyoto, Japan 🇯🇵 — Local Time 09:00:40 PM
It may be surprising to you to learn that I have never driven or even been in a Suzuki Jimny. I did touch one at the Jimny museum though. That was cool.
Another thing my followers should know is that I love the Suzuki Jimny, especially the sandy color (Chiffron Ivory). Alas, they aren't legal in the US.
I have a discord channel where I just post pics of Sandy Jimnys and the caption "Everywhere I look I see his face".
This is the thread version.
I have a discord channel where I just post pics of Sandy Jimnys and the caption "Everywhere I look I see his face".
This is the thread version.
December 18, 2025 at 9:23 AM
It may be surprising to you to learn that I have never driven or even been in a Suzuki Jimny. I did touch one at the Jimny museum though. That was cool.