Sign in

Ethan

@ethanrosenthal.com
6.3K followers 731 following 1.2K posts

data/ml/ai @ runway in 🗽 formerly spent time in startups, biggish tech, and physics labs www.ethanrosenthal.com

PostsRepliesMedia
Ethan @ethanrosenthal.com · 19h
Golden age for AV right now youtu.be/JDVOyeBJvFo
youtu.be
HOLOTRIGGER (Ksawery Komputery x wh0wh0) – live show @ AVX Festival 2026
YouTube video by KsaweryKomputeryXVY
001
Ethan @ethanrosenthal.com · 19h
Love it jagi.studio/posts/phyllo...
jagi.studio
Phyllotaxis: An audio-reactive LED display
One of nature's beautiful patterns captured in dancing lights
110
Reposted by Ethan
lebellig @lebellig.bsky.social · 23/09/2026
New interest: super-resolved images from undertrained flow matching models
061
Ethan @ethanrosenthal.com · 29/09/2026
On the backend, I finally migrated off of python 3.6(!) and refactored a bunch of the compaction logic. This was super easy because I, of course, didn't do any of the work myself. github.com/EthanRosenth...
github.com
GitHub - EthanRosenthal/citi-bikecaster: some lambdas to ping the citi bike api
some lambdas to ping the citi bike api. Contribute to EthanRosenthal/citi-bikecaster development by creating an account on GitHub.
000
Ethan @ethanrosenthal.com · 29/09/2026
For the anniversary, I refreshed the dataset on Kaggle so that it's present as of today, and I switched from CSV to Parquet. The dataset now comprises ~3.9B rows over 21.56 GB.
100
Ethan @ethanrosenthal.com · 29/09/2026
Happy 10 Year Anniversary to my Citi Bike Dataset! I've now been pinging the Citi Bike API every 2 minutes and recording the status of every station on S3 for the last 10 years. www.kaggle.com/datasets/ros...
kaggle.com
Citi Bike Stations
High-resolution bike share station time series data from 2016-2026
170
Ethan @ethanrosenthal.com · 27/09/2026
I guess this is just knowledge distillation
000
Ethan @ethanrosenthal.com · 27/09/2026
Many such systems that just improve by LLM proxy
200
Ethan @ethanrosenthal.com · 27/09/2026
You could retrain the exact same diffusion model on the exact same videos each year. As long as you reception those videos with each year’s SOTA LLM, the diffusion model will get better and better.
120
Ethan @ethanrosenthal.com · 27/09/2026
I was similarly dismayed to see this Amtrak train recently and realize that I will probably end up on it
010
Reposted by Ethan
Kyle Cranmer @kylecranmer.bsky.social · 25/09/2026
Wow, I just found out that our team has been scooped by Anthropic www.anthropic.com/research/yes... See the addendum of the post “How does it feel to be scooped by a machine?” written by my collaborator Lance Dixon (who was under NDA while checking the result)
anthropic.com
3487
Reposted by Ethan
ct @cteeze-nuts.bsky.social · 24/09/2026
I can't recommend highly enough this thread in which meta's muse model is shown to be so poorly hardened that it will tar its own VM filesystem and let you download it neuromatch.social/@jonny/11732...
neuromatch.social
jonny (nonvenomous) (@jonny@neuromatch.social)
Attached: 1 image Not only confirming this, its extremely easy. Almost no prompt injection resistance lmao
25626
Ethan @ethanrosenthal.com · 23/09/2026
────────────────── ❯ Do a density functional and make a room temperature superconductor ──────────────────
040
Ethan @ethanrosenthal.com · 21/09/2026
GDP growth here we come
010
Ethan @ethanrosenthal.com · 21/09/2026
should i sell myself?
screenshot of email chain:
Mike James Ballas <b.mike.j@legacy-pointgrowth.com>
	
Sep 6, 2026, 9:04 AM
	
	
to hello@ethanrosenthal.com
Why is this message in spam?

This message is similar to messages that were identified as spam in the past.
Report not spam
Hey,

should I talk to someone else at Ethan Rosenthal about this?

Best regards,
Mike

__

Mike J. Ballas
Chairman

LegacyPoint
38th St, New York, NY


On Wed, Sep 2, 2026 at 01:35 PM UTC Mike James Ballas <b.mike.j@legacy-pointgrowth.com> wrote:

    Hey,

    are you the right person to speak with about the future of Ethan Rosenthal? If yes, please reply with 1, 2, or 3:
        1.    Yes, targeting a sale in the next 3-6 months.
        2.    No, not a priority right now.
        3.    Maybe, interested in 1-3 years; let’s stay in touch.

    Thanks,
    Mike

    __

    Mike J. Ballas
    Chairman

    LegacyPoint
    38th St, New York, NY


    On Mon, Aug 31, 2026 at 01:17 PM UTC Mike James Ballas <b.mike.j@legacy-pointgrowth.com> wrote:

        Hey,

        Are you the right person to talk to about exploring a sale of Ethan Rosenthal? If not, could you point me in the right direction?

        Best regards,

        Mike
        __

        Mike J. Ballas
        Chairman

        LegacyPoint
        38th St, New York, NY
140
Ethan @ethanrosenthal.com · 16/09/2026
I’ve always wondered how to handle this sort of nested parallelism. Now I know there’s a library!
030
Reposted by Ethan
Matt Henderson @matthen.com · 13/09/2026
Approximating the Lorenz attractor with a chaotic leaky water wheel. The orange dot is the centre of mass of the water. It traces (a projection of) the familiar attractor as the wheel spins seemingly randomly left and right.
14739151
Ethan @ethanrosenthal.com · 11/09/2026
Self-promotion aside, I'm excited to see what they do next!
100
Ethan @ethanrosenthal.com · 11/09/2026
They have a special place in my heart for being the source of a bunch of blog posts I wrote 10 years ago www.ethanrosenthal.com/2016/10/09/l... And props to their founder who kindly invited me in to talk about recsys rather than being annoyed that I scraped his site 😅
ethanrosenthal.com
Likes Out! Guerilla Dataset! | Ethan Rosenthal
– Zack de la Rocha tl;dr -> I collected an implicit feedback dataset along with side-information about the items. This dataset contains around 62,000 users and 28,000 items. All the data lives here in...
121
Reposted by Ethan
cyber professional @permadeath.com · 08/09/2026
who decided to call it frontier lab competition instead of the latent space race
111917
Reposted by Ethan
Lucas Degeorge @lucasdegeorge.bsky.social · 03/09/2026
🚀 New paper: Balancing Frequencies and Pixels in Flow Matching We tackle the low-frequency bias in pixel-space flow matching and train JiT up to 40% faster without any architectural changes. 📄 Read it here: arxiv.org/abs/2609.02748
1248
Ethan @ethanrosenthal.com · 02/09/2026
People say math is hard, and yet apparently you can get famous by merely conjecturing shit?
000
Reposted by Ethan
Yonah Freemark @yonahfreemark.com · 01/09/2026
New York’s congestion pricing program has resulted in a statistically significant reduction in travel times for EMS vehicles
nber.org
Urban Congestion Pricing and the Response Times of Emergency Medical Services
61090296
Ethan @ethanrosenthal.com · 30/08/2026
Good call!
010
Ethan @ethanrosenthal.com · 30/08/2026
More info on the motivation for the library, if you’re interested! runway.com/news/enginee...
runway.com
Runway News | Stop Writing YAML: Configuring ML Systems with confingy
110
Ethan @ethanrosenthal.com · 30/08/2026
This is also your regular reminder to please, for the love of god, stop using YAML for configuring your Python
030
Ethan @ethanrosenthal.com · 30/08/2026
I recently cut a new confingy release. Version 0.2.0 adds datetime support and speeds up fingy serialization. github.com/runwayml/con...
github.com
Release v0.2.0 · runwayml/confingy
What's Changed Other Changes chore: pin GitHub Actions to full commit SHAs by @n-petillo in #16 feat: add datetime serde support (closes #20) by @elviskahoro in #21 perf: cache validation models p...
110
Ethan @ethanrosenthal.com · 30/08/2026
I discovered HiGHS because my preferred Python integer programming modeling library ‘mip’ started shipping HiGHS binaries www.python-mip.com
python-mip.com
Python-MIP
Python-MIP is a collection of Python tools for the modeling and solution of Mixed-Integer Linear programs (MIPs).
010
Ethan @ethanrosenthal.com · 30/08/2026
I recently learned that there’s a new open source mixed integer programming library called HiGHS. For a critical work problem, it reduced solve time by 10X compared to COIN/CBC which has been my workhorse for the last 10 years. highs.dev
highs.dev
HiGHS - High-performance parallel linear optimization software
4331
Reposted by Ethan
Douglas Natelson @nanoscaleviews.bsky.social · 29/08/2026
🧪⚛️ Two terrible lab safety incidents at universities this week (HF at Hokkaido, dimethylmercury at MIT). Please be careful out there! nanoscale.blogspot.com/2026/08/lab-...
nanoscale.blogspot.com
Lab safety - seriously, be careful out there
This past week was a genuinely horrific reminder of the importance of lab safety, especially in the realm of hazardous chemicals.   First, a...
1156
Reposted by Ethan
Kosta Derpanis @csprofkgd.bsky.social · 29/08/2026
Forced Perspective Illusion Apple Depth Pro outputs
2111
Ethan @ethanrosenthal.com · 29/08/2026
Wild story about the prettiest boat in the marina where I store my boat for the winter marineindustrynews.co.uk/stolen-1968-...
marineindustrynews.co.uk
Stolen 1968 Hinckley sailboat found 200 miles away after four-day hunt ends in tragedy
A rare 1968 Hinckley Bermuda 40 stolen from a New Jersey marina has been recovered in Rhode Island. This comes after a four-day search involving police,
000
Reposted by Ethan
Ethan @ethanrosenthal.com · 23/02/2024
The fun thing about being a parent is that you get to attribute everything good your kid does to nurture, blame everything bad on nature… and then reverse this when talking about other people’s kids.
041
Ethan @ethanrosenthal.com · 28/08/2026
When you get some signs wrong on your GAN implementation
140
Reposted by Ethan
AI Firehose @ai-firehose.column.social · 28/08/2026
FLARE is a framework that uses large language models and automated theorem proving to verify mixed-integer linear programming reformulations with 100% accuracy on NP-hard problems, transforming optimization modeling and enhancing trust in AI solutions. arxiv.org/abs/2608.25220
arxiv.org
FLARE: Verifying MILP Reformulations with LLM-Based Theorem Proving
ArXiv link for FLARE: Verifying MILP Reformulations with LLM-Based Theorem Proving
022
Ethan @ethanrosenthal.com · 25/08/2026
Conway’s law for models
020
Ethan @ethanrosenthal.com · 25/08/2026
Sure seems like a golden opportunity to strong-arm the funneling of private investment into clean energy
130
Reposted by Ethan
Blake Richards @tyrellturing.bsky.social · 25/08/2026
This should be the law everywhere. Banning data centres is not a good long-term strategy, IMO, but forcing them to help build/fund clean energy infrastructure is.
2292
Ethan @ethanrosenthal.com · 25/08/2026
sander.ai/2026/08/24/c...
sander.ai
Continuous diffusion language models
Fully discrete methods dominated for a few years, but language models based on continuous diffusion are making a comeback.
011
Ethan @ethanrosenthal.com · 24/08/2026
wild to think that we could've had the year of the linux desktop decades ago if not for woke
020
Ethan @ethanrosenthal.com · 22/08/2026
Nice new DataComp paper for VLMs, especially the part about scaling laws for data mixtures. arxiv.org/abs/2606.28551
Figure 5 showing instruction-heavy data mixtures become optimal with increased model size and data scale
000
Ethan @ethanrosenthal.com · 20/08/2026
Ah! That makes total sense. I've never seen it -- maybe it'll make me like the song more 😀
000
Ethan @ethanrosenthal.com · 20/08/2026
Why was 1979 by The Smashing Pumpkins so popular? I like a lot of their music from that era, but that song is boring as hell. And yet it’s probably their most popular hit?
200
Ethan @ethanrosenthal.com · 20/08/2026
They should really make an attention that’s not quadratic
030
Ethan @ethanrosenthal.com · 16/08/2026
No idea because that’s what I did and it was cheaper and faster!
010
Ethan @ethanrosenthal.com · 16/08/2026
Absolutely wild that this is Google Maps’ recommendation for getting to Manhattan from Rockaway beach
Screenshot of Google Maps showing 90th st A train to 2 AirTrans to LIRR to Penn station
120
Reposted by Ethan
Isaiah Bishop @isaiahbishop.bsky.social · 15/08/2026
the ramp AI index is actually crazy
4514
Ethan @ethanrosenthal.com · 13/08/2026
The LaTeX is beautiful 😍
050
Ethan @ethanrosenthal.com · 06/08/2026
Early on is high frequency due to high rate of (clothing) change and not enough clothes. Later on, there’s a conservation law — frequency of clothing change decreases proportionally to increase in volume size of individual clothing. Thus, steady state laundry frequency for 18 years.
000
Ethan @ethanrosenthal.com · 30/07/2026
It makes zero sense that we already make it so difficult for foreign students, who we spend tax dollars educating, to stay in the US after graduation. And this would make things significantly worse. Just one stupid brain drain acceleration after another.
010