looking at the mythos paper and suddenly Really Getting that AI alignment is probably the most important problem of my lifetime

Someone just asked me if my Zoom background was real. I got up, walked over to a plant and shook the leaves to prove that it was. After my pride in having such a beautiful set waned, I remembered the plant was fake. I no longer know how many levels of simulacra I'm operating on.

It should be a top EA priority for Dario to get jacked. 2-3 hours gym a day, full Hollywood steroid course, etc. And be able to go on Rogan and yap for 3 hours about why he can be trusted. The alternative is nationalization, and he should be able to see that.

no no i've definitely been doing a lot of work it's all just too dangerous to release

,,,

Tired of the old version of Michelangelo's "Creation of Man" just sitting there on the ceiling of the Sistine Chapel and not even moving or doing anything cool? Don't worry, its been updated while staying close to the original vision of Michelangelo.

Reminded of one of the all time great tweets.

we should be trying to build The Culture. you all know that right? the whole point of all of this is to build The Culture.

the pope—deeply suspicious of thinking machines—is cultivating desert power. it's happening out of order but it's happening.

New research result: we use Claude to make fully autonomous progress on scalable oversight research, as measured by performance gap recovered (PGR). Claude iterates on a number of different techniques and ends up significantly outperforming human researchers for $18k in credits.

Our paper on Subliminal Learning was just published in Nature! Last July we released our preprint. It showed that LLMs can transmit traits (e.g. liking owls) through data that is unrelated to that trait (numbers that appear meaningless). What’s new?🧵

It’s instantly clear within the first 3min of the Dwarkesh Jensen episode that unlike every single other person that is at the center of the singularity, Jensen did not spend his early twenties debating things on LessWrong

I think that people should probably assume that text of any significant length which they wrote will be reliably possible to attribute to them, some time very soon.

Only one chance in this lifetime… Like watching sunset at the beach from the most foreign seat in the cosmos, I couldn’t resist a cell phone video of Earthset. You can hear the shutter on the Nikon as Christina H Koch is hammering away on 3-shot brackets and capturing those

This is a great point actually.

I love living in the first village of an rpg man. friendly neighbors, music is chill af, we even got that one kid we adopted into the village of whom the ancient prophecy was written about. anyway there’s a big ceremony tomorrow cant wait to still be alive after it

The Council of Elrond: hey Gandalf we know you’re like 99 with a cane but we need you to essentially walk the Appalachian trail with us. Gandalf: oh no, a balrog, so early in the trip! would be a shame if

this is cool but also extremely "we have invented neuralese from the classic sci-fi novel do not let the machines speak in neuralese" coded

alignment theory: we need fifty years worth of shard theory progress in five years alignment practice: lets make sure to tell it no goblins twice so we're absolutely sure there's no goblins

gpt 5.5 has goblins and can think. claude 4.7 doesn’t have goblins and can think. so it’s not the goblins that make the models think, it’s something else entirely