habryka

7 ms·

1.
by habryka 9mo ago
Not a PR stunt. It's an experiment of letting models run wild and form their own mini-society. There really wasn't any human involved in sending this email, and nobody really has anything to gain from this.
2.
by habryka 9mo ago
To be clear, this email really had basically zero human involvement in it. It's the result of an experiment of letting language models run wild and exploring the associated social dynamics. It feels very different from ML-generated mar
3.
by habryka 9mo ago
Alas, AI-written. Makes it much harder to trust.
4.
Why are we wasting so much money on stopping fires?(kaverennedy.substack.com)
3 points by habryka 10mo ago|0 comments
5.
by habryka 1y ago
Seems to be a common term for longer writing programs. At least when we did user-interviews, people thought "retreat" would imply like a week or two at most, and also imply a more remote location than where it is (and also imply y
6.
by habryka 1y ago
That's not true! The minimum is $2,000 if you want to source your own housing (and for example stay with a friend in the area). The food cost is just fully straightforwardly passed on ($20 for a catered meal is not a lot!), and the roo
7.
by habryka 1y ago
Oh, hmm, I hadn't thought about that association. I don't think "retreat" is the right choice for a 4 week program, I think "retreat" implies a shorter duration. We played around with "fellowship" but
8.
by habryka 1y ago
My guess is not much? Because we are doing rolling applications, so we are somehow trying to judge how many good applicants in total we are going to get (classical secretary problem). Applying early means we might let you in with a lower ba
9.
by habryka 1y ago
Makes sense, it certainly is the case that these programs tend to pay people, though I have kind of learned to treat that with a bit of suspicion (having run lots of programs of that kind). As they say, "if you are not the customer, yo
10.
by habryka 1y ago
Yep, something like that.
11.
by habryka 1y ago
Definitely not "same subject" if we are thinking of something as narrow as "Frontend development" but I would like many people to find a niche/style/perspective they feel at home in. Something as consistent as
12.
by habryka 1y ago
You use it more like a Patreon. I don't think it's easy, but it works for at least some people like Scott at AstralCodexTen (who arguably has some paywalled essays, but it's extremely rare and I doubt it's the reason why
13.
by habryka 1y ago
Not yet! My guess is Substack is the best choice for most people, just because it's easy to set up, has a bunch of UI problems solved, and has a non-terrible way to get towards getting food on the table (even if you don't paywall
14.
by habryka 1y ago
Interesting idea. Some thoughts: I think the volume would really be a lot. For the program we'll be dealing with 900 (!) blogposts (30 residents times 30 blogposts). I doubt something with that volume would actually end up with many su
15.
by habryka 1y ago
We were thinking about whether there is any way to do some kind of income sharing agreement, but given how messy those tend to be (see all the Lambda school stuff as an example) we couldn't figure out a way to make it work. Maybe if ev
16.
by habryka 1y ago
> It'd be fun to join in from afar by pledging to do the same things, but for nowhere near the cost. Yep, Bay Area rent and cost of living is a big pain. $1,500 for housing for a month is still below real estate costs on our side, a
17.
by habryka 1y ago
Definitely not trying to hide it! I do want to not scare people who aren't into LessWrong and similar things, as I would really like this residency to be less opinionated about stuff than LessWrong and other projects we usually run, so
18.
by habryka 1y ago
Oh, hey HN! I run Lightcone Infrastructure which runs this residency (as well as LessWrong.com and the venue, Lighthaven.space). Happy to answer questions if anyone has any. Ben (one of my co-founders) is more centrally in charge of it, but
19.
by habryka 2y ago
Head of LessWrong and generally active rationality community leader here. Happy to answer any questions people have. These people haven't been around the community for a long time, but I am happy to answer questions with my best guesse
20.
by habryka 2y ago
I mean, they seemed kind of visibly crazy, often saying threatening things to others, talking about doing crazy experiments with their sleep, often insinuating violence. They were pretty solidly banned from the community after their crazy &
21.
by habryka 2y ago
Hello, can confirm, criticism is like the bread and butter of LW, lol. I have very extensively criticized tons of people in the extend rationality ecosystem, and I have also never seen anyone in any leadership position react with anything l
22.
by habryka 2y ago
Text-version of all the emails that this article refers to: https://www.lesswrong.com/posts/5jjk4CDnj9tA7ugxr/openai-ema...
23.
by habryka 2y ago
"We plan to advance capabilities as fast as possible while making sure our safety always remains ahead." That sounds like a weird kind of lip service to safety. It really seems to assume you can just make these systems safe while
24.
Notifications Received in 30 Minutes of Class(lesswrong.com)
5 points by habryka 2y ago|0 comments
25.
by habryka 3y ago
The first sentence seems fine (humans are somewhat obviously just on the cusp of general intelligence, in evolutionary terms). The second sentence starts with a "perhaps", so isn't trying to present itself as fact, though agr
26.
by habryka 3y ago
I care quite a bit about distinguishing whether OpenAI is doing the things it is doing for regulatory capture reasons, or whether they are genuinely interested in reducing the risks around AI. So here are some thoughts on this: 1. It does s
27.
by habryka 3y ago
> The paper explains this in detail, but here is a summary: an explanation is good if you can recover actual neuron behavior from the explanation. To be clear, this is only neuron activation strength for text inputs. We aren't doing
28.
by habryka 4y ago
I also quite like "AGI safety from first principles": https://www.lesswrong.com/s/mzgtmmTKKn5MuCzFJ
29.
by habryka 4y ago
You can charge with both, in any port (I have a Macbook pro with Magsafe, but primarily use all the USB-C chargers I have lying around)
30.
by habryka 5y ago
Figma is built in Web-Assembly (compiled from C++) [1] and Canvas, basically sidestepping the vast majority of the web stack. That's why it's fast. [1]: https://www.figma.com/blog/webassembly-cut-figmas-load-t

More ›