Ethan
@ethanrosenthal.com
works @ runway in 🗽
www.ethanrosenthal.com
www.ethanrosenthal.com
I’m honored to serve 🫡
November 5, 2025 at 12:47 PM
I’m honored to serve 🫡
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
Imagine paying $10M to live in Hudson Yards
October 15, 2025 at 12:05 AM
Imagine paying $10M to live in Hudson Yards
I wrote this post 8 years ago, but I think now is the time that it’s actually coming to fruition thanks to LLMs and diffusion. We’re in a golden age. www.ethanrosenthal.com/2017/03/20/a...
September 25, 2025 at 11:43 AM
I wrote this post 8 years ago, but I think now is the time that it’s actually coming to fruition thanks to LLMs and diffusion. We’re in a golden age. www.ethanrosenthal.com/2017/03/20/a...
No biggie, just an original Signal Problems tote 💅
September 24, 2025 at 11:31 PM
No biggie, just an original Signal Problems tote 💅
Fun discovery in the basement trash room
September 9, 2025 at 11:39 AM
Fun discovery in the basement trash room
Spotify leaning back into social. This is quite nice: you can see a full history of every song that you’ve shared with somebody, even if you sent them links to songs off of Spotify.
September 3, 2025 at 11:11 PM
Spotify leaning back into social. This is quite nice: you can see a full history of every song that you’ve shared with somebody, even if you sent them links to songs off of Spotify.
⚠️ Warning! Note! Note! ⚠️ Warning!
August 15, 2025 at 3:21 PM
⚠️ Warning! Note! Note! ⚠️ Warning!
Nevertheless, I managed to frankenstein everything together by attaching the VQD106 pipes to the TD-4KP frame. It doesn't fold up as nicely anymore, but it's good enough, especially given that the whole family can sleep while I play.
August 11, 2025 at 2:37 PM
Nevertheless, I managed to frankenstein everything together by attaching the VQD106 pipes to the TD-4KP frame. It doesn't fold up as nicely anymore, but it's good enough, especially given that the whole family can sleep while I play.
I started with a Roland TD-4KP drumset which is small and foldable. It was discontinued 9 years ago, but I managed to find one on ebay. The downside of this drumset is that the pads feel terrible and are also very loud when you hit them.
August 11, 2025 at 2:37 PM
I started with a Roland TD-4KP drumset which is small and foldable. It was discontinued 9 years ago, but I managed to find one on ebay. The downside of this drumset is that the pads feel terrible and are also very loud when you hit them.
🧵 I finally built the world's ultimate NYC electronic drumset:
It folds up small, and it's incredibly quiet.
It folds up small, and it's incredibly quiet.
August 11, 2025 at 2:37 PM
🧵 I finally built the world's ultimate NYC electronic drumset:
It folds up small, and it's incredibly quiet.
It folds up small, and it's incredibly quiet.
Do CS papers have an Author Contribution section? Physics does this. Not sure how much it actually matters, but maybe it’s a step in the right direction for credit attribution for a medium number of authors
August 10, 2025 at 10:16 PM
Do CS papers have an Author Contribution section? Physics does this. Not sure how much it actually matters, but maybe it’s a step in the right direction for credit attribution for a medium number of authors
No idea! I don't know anything about it.
But apparently the answer is Yes, from the K2 paper
But apparently the answer is Yes, from the K2 paper
![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.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:pv6dad2uw2cit2fffq5jly34/bafkreih74fhjwcvzuogvqdealtqapgd4gjqto5dbfypafwzrrs6fcys33a@jpeg)
August 8, 2025 at 3:54 PM
No idea! I don't know anything about it.
But apparently the answer is Yes, from the K2 paper
But apparently the answer is Yes, from the K2 paper
Priest: You make the statue for the Children’s Sculpture Garden?
Artist: Sure did boss, a bad acid trip just like you asked.
Priest: what
Artist: Sure did boss, a bad acid trip just like you asked.
Priest: what
July 20, 2025 at 12:44 AM
Priest: You make the statue for the Children’s Sculpture Garden?
Artist: Sure did boss, a bad acid trip just like you asked.
Priest: what
Artist: Sure did boss, a bad acid trip just like you asked.
Priest: what
Brb I need to use the stress relief scream jar
July 2, 2025 at 12:32 PM
Brb I need to use the stress relief scream jar
It turns out you can ride that dopamine rush for a full year after a miserable winter and spring
June 20, 2025 at 12:53 PM
It turns out you can ride that dopamine rush for a full year after a miserable winter and spring
It's fun when you don't know if he's talking about the potential next NYC mayor or the president
June 13, 2025 at 6:47 PM
It's fun when you don't know if he's talking about the potential next NYC mayor or the president
I posted slides on my website for the two talks I've given this year
Machine Learning Engineering's Arrested Development: www.ethanrosenthal.com/talks/2025-p...
Building a Data Foundation for Multimodal Foundation Models: www.ethanrosenthal.com/talks/2025-d...
Machine Learning Engineering's Arrested Development: www.ethanrosenthal.com/talks/2025-p...
Building a Data Foundation for Multimodal Foundation Models: www.ethanrosenthal.com/talks/2025-d...
June 13, 2025 at 4:41 PM
I posted slides on my website for the two talks I've given this year
Machine Learning Engineering's Arrested Development: www.ethanrosenthal.com/talks/2025-p...
Building a Data Foundation for Multimodal Foundation Models: www.ethanrosenthal.com/talks/2025-d...
Machine Learning Engineering's Arrested Development: www.ethanrosenthal.com/talks/2025-p...
Building a Data Foundation for Multimodal Foundation Models: www.ethanrosenthal.com/talks/2025-d...
Congrats to all the ML Monitoring tools for successfully rebranding as LLM Evals tools
![Hi Ethan,
My name is [REDACTED] and I work for [REDACTED], an expert network firm based in New York. I am currently working with a client who is looking to speak with current/former customers of End-to-end LLM Evaluation Platforms (Comet, Weights & Biases, Vertix AI (Google), Arize AI, WhyLabs, etc.).](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:pv6dad2uw2cit2fffq5jly34/bafkreibpev7gdft6qfoehwj2uoehvv75obqmjrsamdpf3477uczyq422lq@jpeg)
June 10, 2025 at 3:41 PM
Congrats to all the ML Monitoring tools for successfully rebranding as LLM Evals tools
