Keeks

Stress Testing Deliberative Alignment for Anti-Scheming Training — Apollo Research Future AIs might secretly pursue unintended goals — “scheme”. In a collaboration with OpenAI, we tested a training method to reduce existing versions of such behavior. We see major improvements, but ...

frontier language models know situationally when they are being tested, what they might be being tested for, and what artifacts may be visible to their testers. when you tell them that you can’t see their chain of thought reasoning, they know you’re lying.

www.apolloresearch.ai/research/str...

Excited to build this with you all 🫡

This is so fucking incredible. Decades of research finally able to be given to people. 🥹 I love science, malaria can get fucked

backend team after keeping the servers online the last 3 days

painting of first known ska musician

leaving tw1tter like

The day I get 2 invites for bluesky, some guy turns a joke about male ego lifters into a misogynist one that he must virtue signal on

the Twitter vibes have caught up with this app 🫡

So why project misogyny onto a person you don’t know, as if my joke was directed at someone deadlifting over 4x their body weight (more than enough to qualify for OPs gate)?

Why didn’t you say “I didn’t like your joke, so it must suck” off jump? Less malicious than projecting bad faith and virtue signaling

Never fails

every. single. time.

The goofiness about this attempt at a dunk is that most people treat sumo as easier than conventional, because people read “put your feet wider apart” as “compromise on the form so much that you only pull the bar 1/3 of the distance off the floor vs a regular DL”

or, I have a higher ROM because I focused on strengthening my weak hips, and have high regular and sumo dead lift numbers?

but thank you for reading bad faith into a joke. If anyone was worried that this would differ too much from Twitter, you’ve assuaged those fears

A revolutionary and a scholar

Sumo or actual deadlift

You’re welcome to join 🫂

Happy Mother’s Day, Respectfully 🫂

what if we kissed at the intersection of low self esteem and blinding narcissism

unironically me while playing go

A number of others have said it, but since no meta vr device tracks feet they’d have to guess where your feet are. If they guessed poorly it rep’s ^^ compute but ⏬⏬ immersion.

VRchat is really good, but the std of quality for meta is a lot higher, w/ lower error margin, than competitors

sooo would a scientist who has written a book be willing to briefly chat with me on twitter DMs about book agents? please reply to this if you are! ~things are happening~ but I don’t know anything about how any of this works, would really appreciate talking to someone with more experience

when hetero men skew towards emotional intelligence, the dating pool will be in shambles

Imagine a straight dudebro telling a girl “You’re gorgeous and charismatic, but you have the emotional intelligence of a chocolate doorknob, and I’m allergic to chocolate”

its me, I’m the dudebro

be careful what you wish for

gm

gm

Minas Tirith at Dawn; Hidden City Gondolin; Approaching Edoras, Ted Nasmith

🍨

Hold a funeral for the limerence, grieve, keep it pushing

the same person invented the aeropress and the frisbee and i think that's beautiful