Fundamentally whether a state permits AVs today or not, it's just delaying things a few years in any direction

God these PhD candidates are all so talented we should have 2x more graduate students

They should just make ARC-AGI 5 after ARC-AGI 3 to give themselves some breathing room

wish there could be like the equivalent of knitting or crafting circles where reasonable folks who are reasonably skeptical and tired of ai hype can get together and build little better futures together

RLC 2026 Call for Workshops is up! This is the opportunity to bring the attention of the community to the topics that you find the most important in RL!

An internet pile-on does wind up changing how you view the world a bit. You realize there’s some subset, unclear how large, that is willing to viciously go after an idea of you that exists in their heads

This isn’t to induce controversy but often times making a tool underperform is a cope that lets you avoid the discomfort of how much things will change.

We opened up applications for the Brown AI Policy Summer School! Please share with any computing or computational social science students who want to engage substantively with policymaking in the United States: cntr.brown.edu/summer-school. Deadline March 27th!!! Funding available!

Check out the Call for Workshops at RLC this year. There is still nearly a month before the deadline, we are looking forward to your proposals!

Exciting news... Percepta is growing into doing critical industry transformations in Europe, and that means we're expanding to the UK! If you're an Applied AI engineer looking to work in the UK check it out: jobs.ashbyhq.com/percepta/55a...

Pretty much all the work I'm proud of has come out of being compute bounded so I find tech prognostications about the death of academia due to lacking compute to be pretty odd. You need enough compute to run the minimal experiments (~4 GPUs instantly), but beyond that it's a creative constraint

[1/n] Just wrapped up 7 months interning with @pcastr.bsky.social at Google DeepMind and I'm so excited to share our work: arxiv.org/abs/2602.10324. TLDR: We used LLM-powered program synthesis to automatically model and discover differences between human and LLM strategic behavior

It's a golden age for writing blog posts about how it's a golden age. Who will be the first to draw attention with their post about automating their research. The race is on

Excited to launch Principia, a nonprofit research organisation at the intersection of deep learning theory and AI safety. Our goal is to develop theory for modern machine learning systems that can help us understand complex network behaviors, including those critical for AI safety and alignment. 1

My top educational preference would be for all multi-media to have integrated active feedback and quizzing. Now with LLMs, anything you touch online can instantly also have a quiz to practice retrieval.

🚨 🔬 PhD positions at Google DeepMind in France 🇫🇷 We are advertising Master Level Intern positions at Google DeepMind within our Frontier AI Unit. These could lead to co-advised PhD positions with Google DeepMind and French academic institutions. job-boards.greenhouse.io/deepmind/job...

It’s kind of interesting that so many faculty left twitter but many graduate students did not. I am not sure what the underlying cause there is

Want to learn about the latest in robotics research? Like, the real cutting edge, right from researchers? In addition to our youtube channel, I post notes and information about new RoboPapers episodes on Substack here: robopapers.substack.com

I'm very sad to have learned today that Joe Halpern passed away. Joe was a giant who knew no scientific boundaries. He loved science with a contagious, child-like enthusiasm. He was wonderful and I'm so grateful that I got to learn from him. Thank you Joe 🙏 www.bangsfuneralhome.com/obituaries/j...

Starting the book thread late: I loved "Ducks", though it is weird to speak of love for a book that is mostly so sad. I've been reading Kate Beaton's comics as long as I can remember and she has a way of getting something poignant out of the contrast with her "simple" art. Highly recommend.

Btw the technical term for AI bros is a tensorbro

We need to raise the bar on research code right now. 1) documentation and tests are dead simple now. 2) creating benchmarks integrating across multiple implementations 3) have agents double check your work / fix broken tests 4) fix outstanding bugs in major scientific packages

The skills, such as they are, are text files which you can copy so there’s no moat there. Knowing the edges of your tool is normally systems thinking. None of this is particularly AI specific.

All the "you need to learn AI skills or you'll get left behind" things are patently nonsense. It's easy to use and only becomes easier to use over time. If there's skill it's in knowing what it does well and what is does poorly

The workshops are maybe the best part of RLC. Bring us your workshops that could never happen anywhere else!

🚀 Excited to share REPPO, a new on-policy RL agent! TL;DR: Replace PPO with REPPO for fewer hyperparameter headaches and more robust training. REPPO, led by @cvoelcker.bsky.social, will be presented at ICLR 2026. How does it work? 🧵👇

It's crazy that there's a big ML paper -> 100 million+ raise pipeline now. The incentives have never been worse

3 weeks to get your RLC papers in shape! And to get the word out to those who have yet to experience an RLC review process.

Proud to share our latest work, accepted at @iclr-conf.bsky.social 2026: APPLE! 🍎 TL;DR: APPLE is a novel reinforcement learning framework for solving active perception problems. #ICLR2026 #Robotics #MachineLearning #ActivePerception #RL @ias-tudarmstadt.bsky.social

well, if we are getting accelerationism on fake videos at least we should get good tooling to establish the provenance of videos too

Hiring a postdoc for the Normativity Lab at Johns Hopkins (2026 start). Looking for multiagent systems expertise (RL/generative agents) + interdisciplinary background in AI and cognitive science/econ/cultural evolution. apply.interfolio.com/177701

Waking up every morning, grabbing some coffee, moving some AI goalposts, brushing my teeth

One really funny thing is watching obviously intense people try to pretend they want to do small talk

To help calibrate on the state of AI, it has been weeks since I did a boring "edit this doc to satisfy some administrivia" task as I now automate these. That's productivity growth, and it will have some consequences

Our grad-level "Deep Learning" course (MIT's 6.7960) is now freely available online through OpenCourseWare: ocw.mit.edu/courses/6-79... Lecture videos, psets, and readings are all provided. Had a lot of fun teaching this with @sarameghanbeery.bsky.social and @jeremybernste.in!

Aliens could land and some folks would be too irony poisoned to respond with anything but snark

Should we use a target network in deep value-based RL?🤔 The answer has always been YES or NO, as there are pros and cons. @iclr-conf.bsky.social, I will present iS-QN, a method that lies in between this binary view, collecting the pros while reducing the cons🚀

There is so much barbarity coming out of the government every day that there simply can’t be a straightforward return to normalcy on the other side of this. Someone must be accountable

Let us make a refinement of this claim. The claim is “do not use AI to obscure the level of care that went into your work”

Let us make a refinement of this claim. The claim is “do not use AI to obscure the level of effort that went into your writing.” That’s where the dishonesty lives, not in the use of the tool

This is a very vulnerable and important essay, pointing to the fact that there are so many ways to contribute to computer science research

Well, I went ahead and wrote it. An attempt to work through the discomfort people feel with AI agents on social media by reframing it as an aesthetic problem.

vickiboykis.com/2025/10/20/i... Lovely recent essay on it

Think that it's dishonest to not tell someone when your writing is AI written, the same position I hold on ghost writing

As ever @daphne-cornelisse.bsky.social's figures take every bit of writing to another level

Most IL work in self-driving focuses on the WOSAC score; Daphne's post highlight some critical nuances but one I want to highlight is that a perfect policy, one that never crashed, would do extremely poorly on WOSAC leaderboards open.substack.com/pub/daphneco...

A funny thing that the use of AI-generated em dashes has clued a lot of people into the value of the em dash and caused its human frequency to go up too

help we built a client and now it is being used to bully us

The fact that Claude doesn't know about Claude features is a pretty interesting missing strange loop

@mariaa.bsky.social and I have been building a custom client that we'd like to make public in the near-term BUT we want to make sure we've gotten the security right. Anyone with some expertise open to helping us out a bit?