Ethan

@ethanrosenthal.com

6141 Followers 723 Following

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

Me drumming on an electronic drumset to midnight at mcguffy's
Plot of pm2.5 over time in my living room showing a peak of 100 micrograms per cubic meter
Screenshot of citi bike app showing many empty stations
Metro sign showing 1/2 minute to the next train
“Taste NY” menu at terminal 1 with a “NY Dog” for $20.99
Zoom in screenshot of Strava map showing detour to Amsterdam Avenue at 109th ST
Strava screenshot showing a line from the top of manahattan to the bottom, 16.01 miles, 4h 31m
Screenshot that says KIMI K2 0905 SAYS < 6/8 > MACHINE LEARNING ENGINEER AND DATA SCIENTIST KNOWN FOR WRITING ABOUT ML, PYTHON, AND FOOD OPTIMIZATION AT ETHANROSENTHAL.COM.
Screenshot of article saying “domain expertise has always been the real moat”
Screenshot of a Google search showing a mix of hockey Stanley cup and Stanley tumbler results
LIRR service is suspended A Long Island Rail Road strike is in effect and service is suspended. Get details on shuttle bus service, accessible travel alternatives, and more. Long Island Rail Road Updated May 16, 2026 12:10 p.m.
Signs in a children’s library saying Feelings, Potty, and Vehicles
teenage mutant ninja turtles logo with the words "BURGER KING HAVE IT YOUR WAY"
Screenshot of Amazon product pages for insulated pizza delivery bags
Me with safety goggle, mask, and drill under a boat on land with a propeller behind me
Screenshot of citi bike stats: 2242 rides, 2299 miles, 347 hours
claude code printout showing ● 35/281, about 12%. Let me stop polling and wait for the completion notification. ~8 more minutes. Read 5 files (ctrl+o to expand) ● 45/281 (~16%) at 1:36. Picking up speed. Let me wait a bit longer. Read 3 files (ctrl+o to expand) ● 51/281 (~18%) at 1:47. About 8 more minutes. I'll wait for the completion notification. Read 5 files (ctrl+o to expand) ● 60/281 at 2:05. ~7 minutes remaining. Let me space out checks more. Read 7 files (ctrl+o to expand) ● 70/281 (25%) at 2:28. On track to finish in ~6 more minutes. Read 8 files (ctrl+o to expand) ● 89/281 at 2:55. Speeding up. ~4 more minutes. Read 4 files (ctrl+o to expand) ● 100/281 at 3:09. Now at a faster pace (~1.4s/it). ~4 more minutes. Read 7 files (ctrl+o to expand) ● 111/281 (40%) at 3:35. About 6 minutes left.
Job posting at Mercer for STEM Professor at $85/hour
Good challenge - let me verify rather than speculate.
Fair point. Let me check the actual row count
A subway ad for ketamc.com
Hi Ethan Nannernest caught my eye, specifically how you used computational geometry to optimize banana coverage on peanut butter sandwiches. The way you approached that optimization problem shows you think about algorithmic efficiency in practical applications. I'm working with a seed stage company backed by CRV, XYZ, and the founders of MongoDB & KAYAK on a Software Engineer role. They're building AI-powered financial pipelines for healthcare practices, paying up to $275K + significant equity, based in NYC (onsite). Given your Python expertise, computational geometry background, and CI/CD experience shown in your repos, you'd be solving similar optimization and pipeline challenges at scale for healthcare financial data.
screenshot of 3 git commits with claude + my pic on the 1st and 3rd and just me on the 2nd that says 'ruff' as the message
Cat in the hat hats from the 90s
screenshot of a legend saying PENDING_NODE_ASSIGNMENT RUNNING FINISHED
A Krispy Kreme box that says crocs
A grill and a table on a balcony with a light dusting of snow
Delivery fee This fee covers delivery-related costs for the restaurant and is based in part on the delivery location you provided. THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA. Got it
screenshot of a claude conversation that says too many em-dashes You're right--let me fix that:
Weather app showing 10 degrees
a google search for "ray actor" with actors named Ray in the image search and the ray library documentation as the text search
The Syncopation drumming book
Some textbooks
An electronic drumset in the corner of a room next to some djembes, a tv stand, and a zebra print rug
Screenshot of a comment saying Remember kids: congestion pricing is nothing but a tariff on transportation
A screenshot of the front page of hacker news showing a post in the top spot about NYC congestion pricing cutting air pollution by 22% in 6 months
An upside down handle of vodka named Sea Ice emptying into the fresh water tank of a sailboat
2 panel image: top, mamdani and trump next to each other at the white house, bottom, mamdani by himself in the same spot
Side view of a man drumming on an electronic drumset
Screenshot of the post-training section from HunyuanImage 3.0 Technical Report https://www.arxiv.org/abs/2509.23951
Zillow map screenshot with a circle around a 9.9M apartment
Me as Kurt Russel in Escape from New York
In the meantime, while everybody grapples with this new technology and publishes marginal increases in MNIST accuracy, it seems like the old, differential equation-heavy fields are up for grabs for anybody who can suitably aim the deep learning hammer.
Wall with a pipe and knives shoved between the pipe and the wall
screenshot of a google search. searchbar says "eppo". top sponsored result says "Eppo is Acquired. Try Statsig"
screenshot of the boto3 S3 client docs showing many warnings and notes
Top down view of the VQD106 pads on the TD-4KP frame
Top down view of the TD-4KP drumset with black rubber pads.
Side view of an electronic drumset, animal print drum throne with pink sticks, and city window view
Author contributions: A.C.Q. and P.N. performed the ARPES and LEED experiments and analyzed the ARPES data. S.L. and U.S. grew all graphene samples. A.C.Q. performed tight binding calculations. A.C.Q. developed the thermal atomic hopping model and performed calculations with assistance from T.M. P.L.S. and C.G. analyzed LEED data. A.C.Q., P.N., M.M., M.S., S.Z., and G.L. provided technical support and maintenance for the ARPES setup. C.G. and A.D. supervised the project. A.C.Q. and C.G. wrote the manuscript, with input from all authors. A.D. was responsible for overall project direction, planning, and management.
Budget Control It has been widely observed that RL often results in a substantial increase in the length of modelgenerated responses [35, 19]. While longer responses can enable the model to utilize additional test-time compute for improved performance on complex reasoning tasks, the benefits often do not justify its inference cost in non-reasoning domains. To encourage the model to properly distribute inference budget, we enforce a per-sample maximum token budget throughout RL training, where the budget is determined based on the type of task. Responses that exceed this token budget are truncated and assigned a penalty, which incentivizes the model to generate solutions within the specified limit. Empirically, this approach significantly enhances the model’s token efficiency, encouraging concise yet effective solutions across all domains.