Grounding lets vision-language models do more than describe—they can point to where a robot should grasp, which button to click, or which object to track across video frames. Today we're releasing MolmoPoint, a better way for models to point. 🧵

I feel like a lot of early data scientists got pushed towards analytics engineering or machine learning engineering (myself the latter) due to the need to "productionize" their work and probably because engineering skills are more easily measurable. (1/2)

So first, this is almost exactly what the Gmail AI summary of the email he sent said and my reaction was "Lol the AI fucked up!" The AI did not fuck up... 🫠

Shoutout to “DJuboor” for being the first public contributor to confingy! Version 0.1.1 is out now with pathlib serialization support and some small bug fixes github.com/runwayml/con...

So we’re just doing ketamine subway ads now?

Nothing makes me feel older than working on a random silly side project that absolutely would have crushed it on 2018 data twitter and yet is now so wildly out of step with the moment

lol nothing says “practical applications” like a sandwich algorithm

I'm happy to announce confingy, a library that I've been wanting since the start of my tech career. You can pip install it today! (or, better yet, uv add) Blog: www.ethanrosenthal.com/2026/03/10/c... GitHub: github.com/runwayml/con... Docs: runwayml.github.io/confingy/

see, i'm still useful

I’ve been seeing Jamiroquai hats on the street. I don’t know what it means, but it can’t be good.

Uncharacteristically concise rant from me www.ethanrosenthal.com/2026/03/08/r...

gt;dr (great title; didn’t read)

you and me both claude

Blog: “Socialize the Plan” . On AI and Code Review - laughingmeme.org/2026/03/05/s...

the 3 stages of life

Proposal to call STT vibe-coding from your phone “no-scoping”

Exploring the AI Obedience: Why is Generating a Pure Color Image Harder than CyberPunk? Hongyu Li, Kuan Liu, Yuan Chen, Juntao Hu, Huimin Lu, Guanjie Chen, Xue Liu, Guangming Lu, Hong Huang tl;dr: Flux and NanoBanana fail at precise color filling. arxiv.org/abs/2603.00166

NYC apartment amenities that shouldn’t be: Ice maker Garbage disposal Bathroom fan

Finally, we can get some money for AI. Oh wait openai.com/index/scalin...

lol

Exceeds expectations

What smells of bullshit is the fact that you’d expect companies to do more things with their newfound productivity rather than cut staff and maintain the same supposed output.

The perfect crossover doesn’t exist-

Reducing large library import time is a nice task for claude code (or at least it will be until pep-810 lands peps.python.org/pep-0810/)

This data viz is incredible (as per usual for them) pudding.cool/2026/02/wome... It also brought me back to my days at Dia&Co, struggling to algorithmically get sizing and body shape right.

Maybe AI is less about software companies having no moat and more like they’re all now open source companies.

Solid snow progress

That’s the same hospital boat sent, with great fanfare, to NYC during the pandemic (that ended up doing nothing). It’s like when they reuse parts of an old movie set on a new production.

My first time coming across the new NY law requiring this disclosure. If this GrubHub delivery fee only uses my location, then I get it, and the disclosure makes it sound more nefarious than it actually is. If the delivery fee uses other data, like my proclivity to tip, then this is pretty suss.

What is the relationship between memorization and generalization in AI? Is there a fundamental tradeoff? In infinitefaculty.substack.com/p/memorizati... I’ve reviewed some of the evolving perspectives on memorization & generalization in machine learning, from classic perspectives through LLMs.

Merge like nobody’s watching Skip tests like you’ve never SEV’d Deploy when nobody’s awake And live like you’re a 10X engineer

I'm going to be speaking at Py AI on March 10th in SF. The lineup looks pretty great -- please join if you're around! pyai.events

This article is like inverse Gell-Mann Amnesia for me. I had opinions on rent control, which I knew nothing about. I read this article, and it blew my mind with info that I never would have considered. And yet I’ll continue to be a blowhard about other things that I know nothing about

There might be a little too much organizational inductive bias baked into Sonnet 4.6

Live for the TODOs you’ll never remember with the the SEVs you’ll never forget

Blogged: laughingmeme.org/2026/02/15/s...

Or, as written up in SIGBOIVK last year:

just found out claude code has `/insights` aka claude code wrapped it produces an html page with details about your usage and suggestions for using the tool more effectively extremely interesting

$16 bec at jfk ayfkm

Well, this is great; I look forward to the coming wave of private equity firms buying up preschools. arxiv.org/abs/2601.20123

🤨

A rare dose of optimism for me recently (related to the climate no less!) has been listening to the @volts.wtf podcast. I read Electrify a couple years ago, and the podcast reveals how that thesis has been playing out since the book was published. And it turns out things are moving along!

Pour one out for anyone doing a walk of shame this morning

The "This is the most fun I've had..." sentiment follows my experience as well. I'll add: it's the most powerful computers have felt to me since I first learned to code. Now it's easier than ever to make them do what I want them to.

I wonder how many people are in the world now? Oh wow, 8.3B. I remember when it was like 6 something. I wonder how old the universe is now? Oh ok, still 13.8B years old.

Yes, training foundation models is a gold rush, but nobody tells you how much fun the work actually is. Just a lovely mix of research, systems, and scale, from training to inference.

Yeesh what a week to finally watch one battle after another

On the bright side, nobody’s gonna complain that you’re spending too much time writing tests nowadays