a “moron” could also be a group-theoretic structure. “On the symplectic inverses of left-invariant morons”

morad and moroid

watching my motivational system surfing the edge of the indefference point on the hyperbolic discounting line—always pulled back by my good friend, the cost

once i’ve found a core to corrigibility i’ll write an incel manifesto

ai alignment is great bc nobody knows what it should be, so you can do whatever you want and call it alignment.

some people just shouldn’t be given the responsibility to create datasets.

i want to commit violence

AAAAAAAAAAAAAAAAAHHHHHHHHHHHHHHHH ARE YOU FOR REAL

……………Tetlock data uses AMERICAN DATE FORMAT IN A DATASET

lw karma political compass

the cold waste of unknown en-pyee

the words “brier score” are used inconsistently: sometimes they just refer to the MSE of the probability assigned to the correct outcome, sometimes they refer to the MSE of the sum of the probability assigned to the correct and the incorrect outcome.

retconning my inner child

“The geological study of Antarctica has been greatly hindered by nearly all of the continent being permanently covered with a thick layer of ice.”

is the “takeoff” of fundamental physics in the first half of the 20th century mostly due to the industrial revolution?

goal: develop strong opinions about vector fields and where to stick wires into people’s heads.

even if cause X came along now, people wouldn’t be able to update towards it within a year.

it’s funny, conditional on AGI (and perhaps also WBE?) not doing us in, i’m pretty bullish on this century. bio seems much less of a problem, and everything else is basically a-okay, especially with people becoming richer and needing to fight less. most other collapse narratives sound pretty unlikely (though prepping is sitll a good idea! you should have three months of food & water at home)

if aliens exists and UAPs are caused by them, what implication does that have for AI safety?

hoooly crap the tino rangatiratanga is a good flag

closest pairs of points are not necessarily reflexive. there are some points in euclidean space that are just fundamentally abandoned

“perception is all you need” — Rosenblatt, 1958

it’d be deeply ironic if MLPs, sufficiently scaled, are enough for AGI

A machine-executable plan to install a self-supervised reinforcement learning system on all instances of the spot robot from boston dynamics, described step by step | posted on art station

our next arecibo message should probably include the orders of the sporadic groups, just to show that we’re somewhat advanced as a species.

Bayesian jihad

r strategy poster by day, K strategy poster by night

with resolving inconsistent graph preferences, does the portion of non-uniquely resolvable preferences shrink with the number of options? in other words: for the set 𝓖_n of all graphs with n nodes, and the set U(𝓖_n) of those graphs with a unique path-graph with a smallest graph-edit distance, is |U(𝓖_n)|/𝓖n<|U(𝓖{n+1})|/|𝓖_{n+1}|?

Logarithmic well-being views might suffer from an odd problem: They are, in some sense, too obvious, so that everyone says “yeah, that seems true” and goes on doing exactly what they did before.

“AI” refers to the observation that computers, over time, can do more & more things

The name niplav is based on a roof tile of the University of Maryland, which matches the sequence number and neuroscience thread index, which is how neuroscience researchers manage their data.

“Dodecahedron Assemblies was the publication I chose to read on my way to heaven”—apparently Uriel has managed to gain short contact via the Gato agent.

I really look forward to the next AI winter! We’ll all cozy up inside by the fireside in thick markov blankets, everyone finds a new agent foundations paper in their stocking, and we can listen to uncle Paul have his discussion with uncle Eliezer when the whole family re-unites…

niplav yushtun, ta faret ne digal.

longevity research is trying to move an outer loss signal into the inner optimizer, we’ll have to see whether that one is strong enough to take up the burden

knuckle tatoos: “nplv site”, “ttra spce”, “less wrng”

The Lightgirl of the Gate

DALL·E 2 prompts: The Gateboss of the Girlkeep

Have you considered that removing Occam’s razor from your worldview makes the worldview simpler?

the buddha vector is calculated by finding the path in the space of possible attractors that maximizes the number of attractors it could fall into with an ε perturbation, but falls into none.

vibe computing

Don Knuth really really wants to write a long site and not books. I mean…just take a look at them!

God exists, and she is a woman 😏

why am i just now finding out about examine.com this is pretty cool

in retrospect, i do remember expecting some message about “yes, you passed the test, you were actually living in a lie”, but i think that died when nobody gave me a bad grade for taking too long to become vegetarian. maybe veganism…

we can also make value of perfect information a shorter word, vopi. or voi for just value of information.

induction, abduction, deduction, transduction, traduction, seduction, adduction, eduction, reduction, contraduction

i think eliezer wanted micatwoa to take place in a world that was even more insane than here for us to more clearly see what’s wrong with our world, and the only way he could come up to accomplish that was to make the story take place in actual hell.

“i’ll stop reading at 1:30” this is your internal agents coordinating

saying “somebody should do X” repeatedly while apparently also not doing that thing (while also not being blocked by anything, such as doing something else that’s also important) ought to be discouraged.