An important point about this site is that if you have deep MDP questions, instead of asking an influencer on twitter, you can ask @martypute.bsky.social which is uh, a serious step up

There are like 26 ML researchers actively on this site but they're all really great so it works out

Literally every time I ask my students for help on something, it just instantly becomes more fun to do. A lesson that needs constant re-learning

uv is now stable enough that I'm switching my course over to using it. No more conda, we will create a generation of uv evangelicals

There’s a place in NYC that serves an elotes chaat. America is unstoppable

People who are pro unmoderated speech will often ideologically say "bad views can only be combated by making them visible." Opponents will often say the reverse. What's the best available literature on which way the effect actually goes?

People frequently come here, say it's boring, and peace out. There are easy ways to make it quite interesting, they just take a bit of thought.

Policy churn: maybe epsilon greedy doesn’t matter that much because the Q value argmax action changes constantly arxiv.org/abs/2206.00730

It’s Saturday post papers

If anyone has anything about Slavic paganism in English, please give it to me for the love of the gods 😩

Models under which leaving twitter for bluesky is net good (and the reverse). All are hypotheses. 1. Forcing people who disagree to see their disagreement in tweet form exclusively deepens divisions. Alternate algorithms, which bluesky enables, might avoid this. 2. Algorithmic bias: twitter…

Meta-world is still very much the go to for multi task and meta-RL research but it has some real failings that Reggie fixes up here!

The people on this site crave papers. You can post a random excellent paper from 1978 on any topic and it'll blow up a little bit. You can fill the gap!

It's odd that so little in the educational system discusses the how of learning. How do you know when you've understood a concept? What's the point of doing exercises? What's being productively stuck? Why should one read a textbook? And yet these prior beliefs seem to be what determines outcomes.

goodbye I have enjoyed doing ML but it is now time to pursue a career in my true passion, living as a hermit in the woods

Simulation theory is real

Our NeurIPS submission arxiv.org/abs/2502.08938 did not get in, but it's one of my favorite papers and I think one of the better papers we've ever put out so I want to highlight it

The #1 signature of AI generated code is emojis in print statements

Trying to get onto the right blocklists: - Most of the time I like Ezra Klein - AI can be quite good sometimes - Very often, markets are effective solutions to problems - Building more housing, not necessarily affordable, is necessary for lowering housing prices

Bluesky challenge of the day is to find a complaint about your NeurIPS reviews that hasn't been done before

Will say again that graze is one of the most interesting value propositions of this site. Take a look!

A student reached out anonymously about this, referencing a situation where there might be costs to dealing with a bully of an advisor. So some nuance: 1. In all things outcomes matter. This is mostly an exhortation to realize the consequences are sometimes less than you think. 1/2

New to Bluesky, so here's my lil' intro: Hi, I'm Isa 👋 part designer, part marketer, part product wrangler 🧩 Built Dora AI and more with the rad team at Dora (dora.run) and 3 seasons of the Lux AI Challenge as CDO 🤩 Here to meet fellow product enthusiasts!

Giving a talk at NYU text as data tomorrow (Thursday) at 11! Catch me there buds

I like every bruce springsteen song but only as covers and I don't know what to do with this information

I want to print it out giant and put it everywhere

“One who feeds a crocodile hoping to be eaten last”

There are very few settings where AI can outperform peak human. But, there are also so many situations where we are constrained by available labor! It's a complement not a substitute.

what this is so great

Luv my new job of “guy who removes LLM cruft from code”

The talk is tonight!

There are now enough good ML conferences that deadline chasing doesn't make any sense. It's a huge relief

I miss machine learning as the buzzword. It promised less

Continually can’t get over that the modern propagandists keep loudly announcing “I am going to do a propaganda” and seemingly get away with it.

A running list of untrue things I see on this website semi-frequently 1) google search is getting worse because they’re putting AI in it (???). As anyone who works there will tell you, search is getting harder because the structure of the internet has changed. 2) there are simple tricks to…

"I mustn't like those things I like or the thing I like predicting machine will predict that I will like those things I like which I will, which vexes me." This is not the sci-fi future I imagined.

I think a year later not much has changed and this post still holds up open.substack.com/pub/emergere...

There is a particular way of speaking that treats its recipient as obviously stupid and it is just so, so grating. I often wonder if they speak to people offline this way

Does anyone have a no politics feed please we are drowning. Paper skygest is good but updates infrequently

AI generated reviews are actually really helpful.... to the author. They frequently flag the misunderstandings a bad reviewer will make

Quoting stuff out of context without the link to dunk on someone is block worthy behavior

I've logged out of my twitter account, I'm not sure I can get back in, and that's the end of that chapter. In many ways I think being able to access it was holding me back from finding new venues to discuss and find things

The current review system is costless. Submitting a bad LLM generated paper generates work for others but no consequences for you. If it gets through, chances are no one reads it but you get the stamp on your CV. Something about that equation needs to change to put more skin in the game

I’m hearing that the admin has asked UCLA to not recognize transgender folks identities. Universities need to plan to reject these requests: transgender folks are our students and colleagues. But above that, it’s just a question of basic respect for how people choose to live their lives

Social media feeds today are optimized for engagement, often leading to misalignment between users' intentions and technology use. In a new paper, we introduce Bonsai, a tool to create feeds based on stated preferences, rather than predicted engagement. arxiv.org/abs/2509.10776

Here is our best thinking about how to make world models. I would apologize for it being a massive 40-page behemoth, but it's worth reading. arxiv.org/pdf/2509.09737

The students I teach unfortunately do not read textbooks. This is a tremendous loss as there's not really an equivalent replacement. Youtube videos, blogs, chatgpt, are supplements but not a replacement for a wonderful textbook

I reference data science at the singularity constantly as it's not just a model of why data science blew up, but also how any field or org can organize itself around data, code sharing, and benchmarks: arxiv.org/abs/2310.00865

feed is all politics today please send good RL papers you've read recently

How can scientists apply efficient #AI tools to cognitive science? And what does it mean for how robots learn in new environments? #KempnerInstitute fellow Wilka Carvalho (@cogscikid.bsky.social l) + collaborators outline a new framework in 3 papers. Read more: bit.ly/47Fk9F9 #NeuroAI