Ethan
@ethanrosenthal.com
data/ml/ai @ runway in 🗽
formerly spent time in startups, big tech, and physics labs
www.ethanrosenthal.com
formerly spent time in startups, big tech, and physics labs
www.ethanrosenthal.com
So do all the WBD engineers get to make Netflix salaries now?
December 5, 2025 at 1:00 PM
So do all the WBD engineers get to make Netflix salaries now?
Reposted by Ethan
State of AI: An Empirical 100 Trillion Token Study with OpenRouter openrouter.ai/state-of-ai ...interesting looking at how different models are used for programming versus.. ahem... "role play"
December 5, 2025 at 10:59 AM
State of AI: An Empirical 100 Trillion Token Study with OpenRouter openrouter.ai/state-of-ai ...interesting looking at how different models are used for programming versus.. ahem... "role play"
Reposted by Ethan
Here are the slides and recordings from our Boston DataFusion Meetup in September:
Youtube: youtu.be/wCAud478Dg8
Slides (pdf): drive.google.com/file/d/18KGH...
Youtube: youtu.be/wCAud478Dg8
Slides (pdf): drive.google.com/file/d/18KGH...
Apache DataFusion Boston Meetup: September 12, 2025
YouTube video by Andrew Lamb
youtu.be
December 4, 2025 at 11:51 AM
Here are the slides and recordings from our Boston DataFusion Meetup in September:
Youtube: youtu.be/wCAud478Dg8
Slides (pdf): drive.google.com/file/d/18KGH...
Youtube: youtu.be/wCAud478Dg8
Slides (pdf): drive.google.com/file/d/18KGH...
The key is to reallocate your spending based on each category’s inflation. The more you drink, the less you’ll be worried about the economy.
Why there never was a Vibecession in one chart.
It’s always been about the prices of essentials.
It’s always been about the prices of essentials.
December 2, 2025 at 12:50 PM
The key is to reallocate your spending based on each category’s inflation. The more you drink, the less you’ll be worried about the economy.
Reposted by Ethan
Unit tests in research code feel like something that's slowing you down until they wind up saving you six months of suffering
December 1, 2025 at 2:52 PM
Unit tests in research code feel like something that's slowing you down until they wind up saving you six months of suffering
Another year, another handle to put her to sleep for the winter
November 30, 2025 at 8:34 PM
Another year, another handle to put her to sleep for the winter
My contribution to The Discourse:
Before writing the quoted blog post, I tried to build the thing that I describe in the post.
I couldn't figure it out.
I've been writing Python professionally for a decade, but it was beyond me.
Before writing the quoted blog post, I tried to build the thing that I describe in the post.
I couldn't figure it out.
I've been writing Python professionally for a decade, but it was beyond me.
New blog post: "Why can't we separate YAML from ML?"
Link: www.ethanrosenthal.com/2024/11/19/y...
This is a very wonky post about configuring training loops for ML models 🧵
Link: www.ethanrosenthal.com/2024/11/19/y...
This is a very wonky post about configuring training loops for ML models 🧵
Why can't we separate YAML from ML? | Ethan Rosenthal
As for the ever popular Python vs R vs Julia vs Scala... It's already been decided and the right answer is YAML. It ever has ML in the name!
😉
[image or embed]
— Alex Gude (@alexgude.com) October 27, ...
www.ethanrosenthal.com
November 30, 2025 at 3:03 PM
My contribution to The Discourse:
Before writing the quoted blog post, I tried to build the thing that I describe in the post.
I couldn't figure it out.
I've been writing Python professionally for a decade, but it was beyond me.
Before writing the quoted blog post, I tried to build the thing that I describe in the post.
I couldn't figure it out.
I've been writing Python professionally for a decade, but it was beyond me.
Reposted by Ethan
✨ #NeurIPS2025 paper: Scalable Feature Learning on Huge Knowledge Graphs for Downstream Machine Learning
Combining contrastive learning and message passing markedly improves features created from embedding graphs, scalable to huge graphs.
It taught us a lot on graph feature learning 👇
1/10
Combining contrastive learning and message passing markedly improves features created from embedding graphs, scalable to huge graphs.
It taught us a lot on graph feature learning 👇
1/10
November 28, 2025 at 3:46 PM
✨ #NeurIPS2025 paper: Scalable Feature Learning on Huge Knowledge Graphs for Downstream Machine Learning
Combining contrastive learning and message passing markedly improves features created from embedding graphs, scalable to huge graphs.
It taught us a lot on graph feature learning 👇
1/10
Combining contrastive learning and message passing markedly improves features created from embedding graphs, scalable to huge graphs.
It taught us a lot on graph feature learning 👇
1/10
Reposted by Ethan
I don't know how to put it eloquently into words but my product manager thought leadership take is:
I'm gonna bet on the company that delivers 100 features customers would like over the company that delivers 10 features they marty cagan'd their brains into believing are the best features
I'm gonna bet on the company that delivers 100 features customers would like over the company that delivers 10 features they marty cagan'd their brains into believing are the best features
November 22, 2025 at 4:49 PM
I don't know how to put it eloquently into words but my product manager thought leadership take is:
I'm gonna bet on the company that delivers 100 features customers would like over the company that delivers 10 features they marty cagan'd their brains into believing are the best features
I'm gonna bet on the company that delivers 100 features customers would like over the company that delivers 10 features they marty cagan'd their brains into believing are the best features
Thank you for choosing Amtrak
(Not that you had a choice)
(Not that you had a choice)
November 22, 2025 at 3:24 PM
Thank you for choosing Amtrak
(Not that you had a choice)
(Not that you had a choice)
I’m feeling this
drumset
drumset
November 19, 2025 at 12:53 AM
I’m feeling this
drumset
drumset
Reposted by Ethan
A model for detecting LLM generated text is under review at #ICLR2026. The website using this model detects one of the reviews for this paper as LLM generated ...
🍿🍿🍿🍿🍿🍿🍿🍿🍿
🍿🍿🍿🍿🍿🍿🍿🍿🍿
(As an aside, it looks like the model EditLens that is used for these scores is itself under review at ICLR and classifies one of its reviews as 100% AI which makes the whole thing very meta)
November 17, 2025 at 1:55 PM
A model for detecting LLM generated text is under review at #ICLR2026. The website using this model detects one of the reviews for this paper as LLM generated ...
🍿🍿🍿🍿🍿🍿🍿🍿🍿
🍿🍿🍿🍿🍿🍿🍿🍿🍿
Reposted by Ethan
I wrote about "chart go up" arguments on social media: varunprajan.github.io/blog/chart-g...
Chart go up
Introduction
varunprajan.github.io
November 16, 2025 at 11:10 PM
I wrote about "chart go up" arguments on social media: varunprajan.github.io/blog/chart-g...
Reposted by Ethan
Fish sorting line.
November 11, 2025 at 1:50 AM
Fish sorting line.
Reposted by Ethan
Breaking: we release a fully synthetic generalist dataset for pretraining, SYNTH and two new SOTA reasoning models exclusively trained on it. Despite having seen only 200 billion tokens, Baguettotron is currently best-in-class in its size range. pleias.fr/blog/blogsyn...
November 10, 2025 at 5:30 PM
Breaking: we release a fully synthetic generalist dataset for pretraining, SYNTH and two new SOTA reasoning models exclusively trained on it. Despite having seen only 200 billion tokens, Baguettotron is currently best-in-class in its size range. pleias.fr/blog/blogsyn...
I’m honored to serve 🫡
November 5, 2025 at 12:47 PM
I’m honored to serve 🫡
“…you’re much better served as a student by getting a degree in an “irrelevant” subject that motivates you to put in real, high-quality work than by shuffling through something more relevant on a surface level doing only the bare minimum to meet the requirements.”
open.substack.com/pub/chadorze...
open.substack.com/pub/chadorze...
College Isn't About Course Content
What matters is learning how to learn
open.substack.com
November 2, 2025 at 1:11 PM
“…you’re much better served as a student by getting a degree in an “irrelevant” subject that motivates you to put in real, high-quality work than by shuffling through something more relevant on a surface level doing only the bare minimum to meet the requirements.”
open.substack.com/pub/chadorze...
open.substack.com/pub/chadorze...
Reposted by Ethan
Latest newsletter (free) covers the newest AI video models, some misc 3d tools, an OCR model bonanza, a few games links including a fun new AI-assisted 2d tile project, my fairytale hunt project, and more... arnicas.substack.com/p/titaa-72-f...
TITAA #72: Fairy Tale Hunters
OCR Bonanza - Real-time Video - Inkwell - Fairytales - Cursed Houses - Training Tips
arnicas.substack.com
November 2, 2025 at 9:08 AM
Latest newsletter (free) covers the newest AI video models, some misc 3d tools, an OCR model bonanza, a few games links including a fun new AI-assisted 2d tile project, my fairytale hunt project, and more... arnicas.substack.com/p/titaa-72-f...
Reposted by Ethan
In astronomy, an analogous effect is the Malmquist bias. "Usually heard in the phrase, 'You fool, you forgot to account for the Malmquist bias.'" -- Ethan Vishniac
en.wikipedia.org/wiki/Malmqui...
en.wikipedia.org/wiki/Malmqui...
October 30, 2025 at 2:06 PM
In astronomy, an analogous effect is the Malmquist bias. "Usually heard in the phrase, 'You fool, you forgot to account for the Malmquist bias.'" -- Ethan Vishniac
en.wikipedia.org/wiki/Malmqui...
en.wikipedia.org/wiki/Malmqui...
Reposted by Ethan
As promised, here's a longer explanation of why you should vote YES on all the NYC ballot initiatives EXCEPT #6
✅ YES on #1
✅ YES on #2
✅ YES on #3
✅ YES on #4
✅ YES on #5
🚫 NO on #6
✅ YES on #1
✅ YES on #2
✅ YES on #3
✅ YES on #4
✅ YES on #5
🚫 NO on #6
I'll write a longer post on why, but here's how I'm voting on the ballot initiatives, and I strongly encourage you to do the same.
✅ YES on #1
✅ YES on #2
✅ YES on #3
✅ YES on #4
✅ YES on #5
🚫 NO on #6
✅ YES on #1
✅ YES on #2
✅ YES on #3
✅ YES on #4
✅ YES on #5
🚫 NO on #6
Go vote! And remember to flip your ballot for the charter amendments!
October 26, 2025 at 6:16 PM
As promised, here's a longer explanation of why you should vote YES on all the NYC ballot initiatives EXCEPT #6
✅ YES on #1
✅ YES on #2
✅ YES on #3
✅ YES on #4
✅ YES on #5
🚫 NO on #6
✅ YES on #1
✅ YES on #2
✅ YES on #3
✅ YES on #4
✅ YES on #5
🚫 NO on #6
(translating Icarus for the modern age)
Oh he'd been cooking alright, but then he was cooked
Oh he'd been cooking alright, but then he was cooked
October 23, 2025 at 3:48 PM
(translating Icarus for the modern age)
Oh he'd been cooking alright, but then he was cooked
Oh he'd been cooking alright, but then he was cooked
Reposted by Ethan
Choudhury and Kim et al., "Accelerating Vision Transformers With Adaptive Patch Sizes"
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
October 22, 2025 at 8:08 PM
Choudhury and Kim et al., "Accelerating Vision Transformers With Adaptive Patch Sizes"
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
Transformer patches don't need to be of uniform size -- choose sizes based on entropy --> faster training/inference. Are scale-spaces gonna make a comeback?
sometimes I feel like I work for the government
October 22, 2025 at 1:07 PM
sometimes I feel like I work for the government