Kabir Kumar, aiplans.org

is there any candidate who's running on a platform to regulate recommendation algorithms?

they will take everything before they kill us and then take more still

The Company Man To get to the campus, I have to walk past the fentanyl zombies.

this is one of the most specifc and funny things I've read in a while: tomasbjartur.substack.com/p/the-compan...

ahhh. that makes sense, i think. i think im still confused as to why the bill change did that, when it just allowed mining when it wasnt allowed before

I'm confused. I thought you were saying that some americans tried removing the law so they could mine the metals needed, but dod stopped them because that would somehow give away ip, right?

>
1) cuz they were doing a huge legal oopsie by handing over military tech IP to the Chinese because there was no other way of making advanced sensors etc

at this point though, surely china is ahead??

though more religions

btw, this is what I feel like with alignment and incentive structures

What the fuck. Why did the DoD do that?? And why has no other country tried to jump on it if they've seen how much of an advantage China gained from it??

guess how this gets funnier, without looking it up

Dwarkesh feels like it's gotten watered down imo

Laudatory definition. Expressing praise and commendation

For those of less erudite persuasion, such as myself:

Shared it with a bright young teammate!

At best, this is addiction, not abuse. But really, I think the ai might just be understanding that the most likely thing to do when given the chance to gamble a lot, is to gamble a lot and when given the chance to gamble more, gamble more.

Evals Guide Some definitions or descriptions will be imprecise. Part of the problem to solve is finding words which more accurately describe what we're looking for. Feedback, comments and questions are very welco...

docs.google.com/document/d/1...
which has eval red teaming components

actually hosting an alignment evals hackathons, where teams are red teaming alignment evals, for this.
also, very cool to say, that someone has made a paper inspired by our previous red teaming approach in hackathons: anloehr.github.io/ai-projects/...

and making a course (very early stage):

Us:
www.lesswrong.com/posts/35vPhn...

it's obv a smaller model??

all of these evals themselves need to actually be thoroughly tested

gm

Congratulations Dr. Edstedt!!

I want them to sound more like robots tbh

what's the sexual connotation???

nooki.me/c/AISafety

affiliating, sure. to actually do stuff and find people to work with though, who are really strongly motivated to actually change the world and willing to say when they're wrong, there's not many better atm.

what about potato waffles tho??

gonna be using 'professor-ass web app' in my lingo from now on

I need to use bsky more because I'm still amazed to see this not erupt into name calling and talking past each other

this is great, thank you!!

really? i don't think it has, tbh