Rendered at 01:48:34 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
gwern 10 minutes ago [-]
> The goal of the game is to give hints to a secret word without giving the same hint as another player; it was shocking how often all the agents playing would give the same hint, even at temperature 1. We resolved this by giving them all “personalities” which were really just topics; for example we’d be telling one agent to think about things in the context of sports, so their clue for “shell” might be “defense” or something, whereas the agent told to be a hippie might say “cancer” for the zodiac’s crab.
Mode-collapse is characteristic of pretty much all chatbots post-davinci-002 and a major challenge to anything creative or game-like.
rtpg 2 hours ago [-]
> This included what was probably the most interesting part, writing the tricky recursive code to implement the match statement.
Back in school I found this book[0], "The Implementation of Functional Programming Languages". Flipping through it was very neat because it kinda demystified the nuts and bolts of implementing the "fancier" functional programming features for me.
Chapter 5 is about implementing pattern matching.
You read this kind of stuff and realize how easy things become when you just tackle a problem with the right mindset and abstractions. And this is a book from 1987! We've always had decent answers for a lot of the stuff.
I think it's interesting to flip through if you're a language implementer, though helps to be comfortable with reading some stuff sideways and not understanding every detail.
Reading this conjured some feelings in me. At various points in my career, I’ve wanted to take a sabbatical and go to RC. Having spent the last year in the agentic coding world, the idea of writing code by hand seems quaint and outdated. Many have spoken of this feeling of loss so I’m not saying anything new here, but for some reason reading about RC in 2026 specifically cemented the idea that I’ll probably never write anything by hand again.
triyambakam 4 hours ago [-]
I always shudder hearing about RC. It feels like both wrong ends of career and hobby intersecting and the whole experience sounds dank and clammy.
dev_dan_2 4 hours ago [-]
I am sorry about the negative feelings and associations you have towards RC. Let's think constructively: What are 1-2 points you would do differently?
azhenley 4 hours ago [-]
Really? Why? Sounds like an ideal way to spend a sabbatical to me!
ykonstant 6 hours ago [-]
Let me guess: the webpage simply contains a link to itself.
rus20376 5 hours ago [-]
This place sounds like an adult daycare.
Has there been a business started based on work done there? Publishable research results? (that second one is actually a pretty low bar, and yet if there is something it is not easily discoverable)
blululu 4 hours ago [-]
Here is an honest good faith answer to a pretty aggressively hostile question:
Adult daycare is almost the opposite of what Recurse center is. Recurse is largely unstructured: people do what inspires them and make their own plans. You will find more structure, hand holding and mandatory time wastes in virtually any business than you will at Recurse Center.
A 12 week batch is not a lot of time to carry out a major research project, but it is generally not intended to treat this as a start up incubator. There are other places for that kind of activity. Recurse is a place to freely learn and explore. There are a few companies that have emerged but they are not huge money machines (most companies are not). There is also some interesting research that is done.
Some things that pop into my head immediately are:
Dan Luu's studies of Motion to Photon Latency in keyboard input [1]
Raph Levien's work on Xi [2], and Andrew Kelly's work on Zig [3].
I'm sure there are other people could dig up more examples, but that really isn't the point. A school's purpose is to produce educated people. Recurse is more of an unschool but the aim is to help people learn and think - it should be judged on those terms first and foremost.
From what I see, this is more of a sabbatical time to just explore and enjoy programming without needing a productive end. I've been considering it myself.
simonw 15 minutes ago [-]
> Has there been a business started based on work done there?
Greg Brockman went through Recurse in Summer of 2015, and co-founded OpenAI in December of 2015.
thomascountz 5 hours ago [-]
I don't get it. Why would you ask if an adult daycare leads to a business or new discoveries? Daycares are for nurturing curiosity and fostering development in a socially engaging environment—much like the Recurse Center—not for publishing papers.
Bjartr 3 hours ago [-]
Even if for the sake of argument its conceded it's a "daycare", what's wrong with that? No one's being forced to go there. It's free for the participants. There's no harm in a group that facilitates people exploring what interests them in a setting where others are doing the same.
saulpw 5 hours ago [-]
You are missing the point. On the line between explore and exploit, RC is basically 100% explore. The point is to learn, and more importantly, to flex the muscle of learning without being taught. And to participate in a community of kind and curious people who are also motivated to do the same.
applfanboysbgon 4 hours ago [-]
The practical goal of Recurse is to make connections. Starting a business or publishing research are explicitly non-goals.
roncesvalles 4 hours ago [-]
The paradox of networking events is nobody who attends "networking events" is worth networking with.
applfanboysbgon 4 hours ago [-]
It's weird to say that about a YCombinator-affiliated event on a YCombinator forum. I'm not especially fond of YC but there's pretty clearly potential to make valuable connections in this sphere.
Mode-collapse is characteristic of pretty much all chatbots post-davinci-002 and a major challenge to anything creative or game-like.
Back in school I found this book[0], "The Implementation of Functional Programming Languages". Flipping through it was very neat because it kinda demystified the nuts and bolts of implementing the "fancier" functional programming features for me.
Chapter 5 is about implementing pattern matching.
You read this kind of stuff and realize how easy things become when you just tackle a problem with the right mindset and abstractions. And this is a book from 1987! We've always had decent answers for a lot of the stuff.
I think it's interesting to flip through if you're a language implementer, though helps to be comfortable with reading some stuff sideways and not understanding every detail.
[0] https://simon.peytonjones.org/assets/pdfs/slpj-book-1987-sea...
Has there been a business started based on work done there? Publishable research results? (that second one is actually a pretty low bar, and yet if there is something it is not easily discoverable)
I'm sure there are other people could dig up more examples, but that really isn't the point. A school's purpose is to produce educated people. Recurse is more of an unschool but the aim is to help people learn and think - it should be judged on those terms first and foremost.
[1] https://danluu.com/input-lag/ [2] https://www.recurse.com/events/localhost-raph-levien [3] https://www.recurse.com/events/localhost-andrew-kelley
Greg Brockman went through Recurse in Summer of 2015, and co-founded OpenAI in December of 2015.
You say that like it's a bad thing.